DEV Community: Frank @ Four-Leaf

Finance data analyst interview questions, and what they test

Frank @ Four-Leaf — Fri, 16 Oct 2026 17:05:52 +0000

Originally published on four-leaf.ai.

Under the Bank Secrecy Act, financial institutions must file reports of cash transactions exceeding $10,000 (daily aggregate amount), according to FinCEN. Read that rule as a query and it is a GROUP BY with a HAVING clause, where the hard part is deciding what the group and the day are. That is the core of finance data analyst interview questions. The SQL is ordinary. The data underneath it is money, and every row carries a rule.

What do finance data analyst interviews actually test?

A finance data analyst interview tests whether you can be trusted with numbers that have to tie out. In most analytics roles, a dashboard that is slightly off gets corrected next sprint. In a bank, a broker-dealer or a lender, a figure that does not reconcile to the general ledger can become an audit finding, and a figure sent to a regulator is a record the firm has to stand behind.

That changes what the interviewer listens for. SQL fluency is the entry ticket and will be tested directly. What separates candidates is whether they ask the questions a finance team asks before trusting a number. Which system is the book of record? Is this table at the transaction level or the position level? Are amounts signed, or is there a separate debit and credit flag? Which time zone defines the business day?

Prepare for four kinds of rounds, in whatever mix the posting suggests. A SQL screen on transaction-shaped data. A domain round on reconciliation and data quality. A regulatory or controls conversation, lighter for fintech product analytics and heavier for risk, compliance and finance teams. And a behavioral round where the stories that land are about catching an error before it went out.

Which SQL questions come up with transaction and ledger data?

Finance SQL questions are mostly window functions, date logic and joins that have to match exactly. Five patterns cover much of what you will see.

Running balances. Given a transactions table, return each account's balance after every transaction. This is SUM(amount) OVER (PARTITION BY account_id ORDER BY posted_at, transaction_id). The tie-breaker on transaction_id matters, because two postings with the same timestamp will otherwise give a nondeterministic balance.

Daily aggregation per customer. Total each customer's cash activity per business day and flag the days above a threshold. The interviewer is watching whether you ask what a business day is, and whether the threshold applies per account or per customer.

As-of joins. Value each position at the price or exchange rate in effect at the time, not the latest one. Expect a correlated subquery, a LATERAL join, or ROW_NUMBER() over prices dated on or before the trade date.

Deduplication. A feed was loaded twice. Find and remove the duplicates without dropping two genuine identical trades. The right answer starts with asking what the natural key is.

Reconciliation. Compare two systems and list every record that is missing on either side or differs in amount. A FULL OUTER JOIN on the shared key with a WHERE clause on nulls and mismatches is the expected shape. The SQL interview guide covers the general mechanics of the live coding round.

Why do interviewers ask about reconciliation and data lineage?

BCBS 239 shows why regulators care about reconciliation and data lineage. The Basel Committee on Banking Supervision wrote, in its Principles for effective risk data aggregation and risk reporting (published 09 January 2013 and known as BCBS 239), that the crisis which began in 2007 revealed many banks, including global systemically important banks, "were unable to aggregate risk exposures and identify concentrations fully, quickly and accurately." Those principles are aimed at banks' risk data and risk reporting.

In the interview this shows up as scenario questions. Two reports show different totals for the same portfolio. Walk me through how you would find out why. A strong answer moves from the outside in. Confirm both reports cover the same date, entity and currency. Check whether one is trade-date and the other settlement-date. Then compare at the lowest common grain until the difference is isolated to specific records.

The follow-up is usually what you would do about it. A good answer names who owns the source system, documents the break, and proposes a check that would catch it next time. Interviewers want to hear that the process changed along with the number.

What regulation should a finance data analyst know before the interview?

A finance data analyst should know which rules shape the data in front of them, without pretending to be a lawyer. Three are worth a working understanding.

The Bank Secrecy Act. FinCEN's summary says the BSA's implementing regulations require financial institutions to keep certain records, file cash transaction reports, and "report suspicious activity that might signify money laundering, tax evasion, or other criminal activities." Analysts on anti-money-laundering teams often build and tune the transaction monitoring rules that feed those reports, so expect questions about structuring patterns and false positive rates.

SEC Rule 17a-4. Broker-dealers must preserve many required records "for a period of not less than 6 years, the first two years in an easily accessible place." The rule's electronic storage provisions require either an audit trail that permits "re-creation of the original record if it is modified or deleted" or storage in a "non-rewriteable, non-erasable format." One way to meet that is to record corrections as new rows rather than overwriting old ones, which is why an interviewer may ask how you would design a correction workflow.

Regulation B. For credit and lending analysts, the CFPB states that Regulation B "protects applicants from discrimination in any aspect of a credit transaction." On April 22, 2026, the CFPB issued a final rule that removed the "effects test" from Regulation B. If you are interviewing for credit model work, knowing that this changed recently signals that you follow the rules your models live under.

Which finance concepts will they expect you to read?

Finance analytics teams expect you to read the vocabulary of the business you are analyzing. You need to know what the columns mean.

For corporate finance and FP&A analytics, that means the three financial statements and how they connect, the difference between accrual and cash accounting, and why revenue recognized this quarter may not match cash collected. The SEC's EDGAR full-text search covers more than 20 years of filings, which makes it a free place to practice reading a real income statement before you are asked to model one.

For trading and investment data, it means positions versus transactions, trade date versus settlement date, realized versus unrealized P&L, and basis points. For lending, it means origination, delinquency buckets, charge-offs and vintage curves. For payments, it means authorization versus capture, chargebacks and interchange.

Pick the set that matches the posting and learn it well. At a lender, settlement conventions matter far less than knowing what makes a loan delinquent.

What behavioral questions do finance analytics teams ask?

Finance analytics teams ask behavioral questions about accuracy under pressure, because the cost of a wrong number is high and the deadlines are fixed. Expect versions of these.

Tell me about a time you found an error in a number someone else had already reported.
Describe a time two data sources disagreed and you had to decide which one to trust.
Tell me about explaining a technical finding to a senior stakeholder who did not want to hear it.
Describe a time you had to deliver an analysis on a close or reporting deadline with incomplete data.

The first one is the most revealing. The interviewer wants to hear how quickly you raised it, who you told, and what you changed so it would not recur. Stories where you quietly fixed someone else's error read as a judgment problem in finance, even when the fix was correct.

All four work in the structure from our guide to behavioral interview preparation. In finance the stakes in the story are audit findings, client statements and regulatory filings, so name them.

What is overrated

Memorizing finance formulas is overrated for analyst roles. Interviewers can tell the difference between a candidate who recites a definition and one who can explain what a number means for the business, and they hire the second. Spending a week on derivative pricing for a payments analytics role is time taken from the SQL and reconciliation questions you will actually be asked.

Firm-specific question lists are also overrated. Running balances, as-of joins and reconciliation repeat across banks, lenders and fintechs because the data has a similar shape.

A five-step prep plan

Write the five SQL patterns from scratch. Running balance, daily aggregate with a threshold, as-of join, deduplication, two-sided reconciliation. Time each one.
Prepare one reconciliation story. A time two numbers disagreed and you found out why, told in under two minutes.
Read one real filing. Pull a 10-K from EDGAR for a company in the sector you are interviewing in and trace one line from the income statement to the cash flow statement.
Learn the rules for your sector. BSA for anti-money-laundering roles, Rule 17a-4 for broker-dealers, Regulation B for lending. One paragraph each, in your own words.
Run the loop out loud. Four-Leaf's AI mock interviews include a data analyst track with a recruiter screen, a hiring manager round, SQL live coding and a business case. Name the employer and paste the posting, and the SQL round's questions are built around that posting's domain.

If you are weighing finance against other regulated analytics work, our guide to healthcare data analyst interview questions covers the same loop on claims data. For the statistics and case questions that sit alongside SQL, see how to prepare for a data science interview.

Where this is heading

We expect finance data work to keep moving toward the controls side, where more of the analyst's job is proving that a number is complete and traceable. The candidates who stand out already think that way in the interview.

They ask where the data came from before they write the query. Interviewers notice, because that question is the job.

About Four-Leaf

Four-Leaf is the all-in-one AI job search assistant covering every stage from application to signed offer: voice mock interviews across 24 roles, resume tailoring, cover letters, salary negotiation, email drafting, AI job search, and LinkedIn optimization. 3-day free trial, no credit card. Pricing: $20/mo or $5 / 5 Day Pass.

The Deloitte interview process, from Deloitte's own recruiting pages

Frank @ Four-Leaf — Fri, 16 Oct 2026 16:44:50 +0000

Key takeaways

Deloitte's US recruiting tips page gives a short, specific answer to how many interviews to expect. Expect two to three rounds of interviews, held by telephone, video, or in person, which may mix behavioral, technical, and case-based interviews depending on the role. The rest of the Deloitte interview process is spelled out across a set of recruiter-written pages, and the clearest message running through them is that Deloitte wants specific examples, said in your own words, without AI help in the room.

What are the stages of the Deloitte interview process?

Deloitte's recruiting tips page lays the US process out in five steps.

Meet us. Deloitte describes a series of structured steps that evaluate candidates on skills, potential, and fit.
Start your search. Deloitte encourages candidates to attend its events and meet its people, and points them to its Explore Your Fit career quiz to explore business areas that may be a good fit.
Apply. Your resume and qualifications are reviewed, and you are contacted if you are selected for an interview.
Interview. A member of the recruiting team schedules an initial interview, usually within one to two weeks. Expect two to three rounds. If you advance, you meet several Deloitte professionals and may go through a mix of behavioral, technical, and case-based interviews, depending on the role.
Offer and onboard. If you are receiving an offer, Deloitte says you will likely have it in writing within one week of your final interview.

One detail on the application step changes how to apply. Deloitte's application FAQs say the recruiting professional compares your skills and experience with the requirements of the role you applied for, as well as any other listed position. A resume that only makes sense for one narrow job description gives that reviewer less to work with.

The footer on every Deloitte page cited here is a reminder that "Deloitte" is a network of member firms that are legally separate and independent entities. Everything in this guide comes from the US firm's pages. If you are applying elsewhere, ask your recruiter which steps apply.

What does the Deloitte behavioral interview test?

Deloitte's behavioral interview tips are direct about what sits behind each question. The page pairs four sample questions with what the recruiter wants to learn.

What the recruiter asks	What they want to know
A time you were juggling a lot of responsibilities	How you balance priorities
A time you saw a problem and took the initiative to correct it	Your problem-solving process
A conflict you had with a team member	How you lead and work with others
A rush project where you lacked the information you needed	How you deal with ambiguity

Deloitte's separate interview preparation page says interviews can help employers evaluate a common set of skills, including leadership, communication, managing ambiguity, and balancing priorities. The two lists overlap without matching, so between them they name six skills: leadership, communication, problem-solving, working with others, managing ambiguity, and balancing priorities. That is a short enough list to prepare one story for each.

Both pages recommend the STAR method (situation, task, action, result), and the behavioral page asks you to refer to the experiences and periods on your resume so the interviewer can connect them to the role. The preparation page adds a warning worth taking literally. It says not to make stories up, because your interviewer will know the difference, especially when they ask follow-up questions.

How do Deloitte case and scenario interviews work?

Deloitte says candidates may be asked to complete a case or scenario interview to demonstrate problem-solving, analytical ability, and strategic and logical thinking. Its case interview tips page says the cases are fictional examples of the types of real client projects Deloitte works on every day.

The page gives a five-step approach for working through one.

Understand the issue, asking clarifying questions as needed.
Identify the underlying assumptions.
Summarize the specific issues and findings.
State your recommendations.
Outline next steps and expected results or impacts.

Two lines on that page matter more than the framework. Deloitte says you are not expected to provide the right answer, but to clearly convey your logic and thought process. And Sophia Fanucchi, a campus recruiting specialist at Deloitte Services LP, puts it more bluntly: "The interview is not about getting the right answer, it's about how well you respond under pressure."

The page also warns about clarifying questions. Ask them, but don't use them as a crutch, because Deloitte is interested in how you connect the dots.

Deloitte points candidates for an entry-level Consultative Offerings position to its case interview tool and candidates for an entry-level Audit & Assurance position to its scenario interview tool. Both are linked from the case tips page and are worth doing before the real thing. Four-Leaf's case interview preparation guide covers the frameworks in more depth.

Can you use AI in a Deloitte interview?

No, unless Deloitte tells you otherwise. Deloitte's guidance on AI in the recruitment process draws the line at the interview itself. Among the uses it does not condone is receiving real-time assistance from AI tools or another person during an interview or assessment unless Deloitte has explicitly permitted it. The page states the consequence plainly: inappropriate use of AI during the recruitment process will negatively impact your application.

The same page is generous about preparation. You may use AI to research Deloitte and the role, brainstorm how to present your experience, refine your resume, and generate sample interview questions so you can practice on your own time. After an interview, Deloitte says you can use AI to reflect on it and generate questions to practice for later rounds.

That split is the useful part. Deloitte is telling candidates where the help belongs, which is before and between interviews, in rehearsal. It is the same distinction Four-Leaf draws between real preparation and cheating: practice that changes what you can do on your own counts, and anything that speaks for you in the room does not.

What mistakes do Deloitte recruiters see most?

Deloitte's common interview mistakes page lists ten, and three of them come with specifics worth planning around.

Claiming skills without examples. Bryce Turkheimer, a campus recruiting specialist at Deloitte Services LP, says a common mistake is candidates who list skills but don't convey them in the interview because they weren't prepared with specific examples.

Rambling. The page says that if it takes you five minutes to answer any one question, you may appear to be struggling to communicate. Answer, then stop and let the interviewer ask the next question.

Arriving late. Deloitte asks candidates to arrive at least 10-15 minutes early, even for a virtual interview.

The video interview tips page adds a permission worth knowing about. You can post your resume, reminders of stories, and notes behind your laptop. Deloitte asks only that you don't script your answers and read them verbatim.

What is overrated in Deloitte interview prep?

Memorizing a polished answer to every question on a list. Deloitte's pages point the other way on almost every count. The case page says you aren't expected to find the right answer. The video page says notes are fine and verbatim scripts are not. The preparation page warns that follow-up questions expose a story that isn't real, and the AI page rules out live help unless Deloitte permits it.

What Deloitte asks for instead is harder to fake and easier to practice. A small set of true stories, each tied to one of the skills its pages name, told concisely enough to leave room for the follow-up. The behavioral page quotes Joel Pearson, a senior campus recruiting specialist at Deloitte Services LP, saying it can be difficult to keep your composure going from one question to the next, "so I'm a big proponent of doing mock interviews."

How to prepare for a Deloitte interview

Ask your recruiter which interview types your role includes (behavioral, technical, case, or scenario) and, as Deloitte's preparation page suggests, for a list of the people you'll be meeting.
Write six stories, one each for leadership, communication, problem-solving, working with others, managing ambiguity, and balancing priorities, in STAR shape with a concrete result.
If you are interviewing for an entry-level Consultative Offerings or Audit & Assurance role, work through Deloitte's own case or scenario tool first.
Practice the five-step case approach out loud, saying your assumptions before your recommendation.
Rehearse your stories out loud and time them. An answer that runs toward five minutes needs cutting, and a voice mock interview is one way to hear that before a Deloitte interviewer does.
Prepare questions about the team and the work. Keyana Dunn, a senior experienced hire recruiting specialist at Deloitte Services LP, says, "You're interviewing an organization just as much as they're interviewing you."

The pattern worth taking away

Deloitte's US recruiting pages quote different recruiters on different stages of the process, and they agree with each other. Use AI to prepare, then walk in alone. Bring real stories, keep them short, and expect a follow-up question to test each one.

Prepare answers that still hold up on the second question.

Originally published on the Four-Leaf blog.

Healthcare data analyst interview questions, and what they test

Frank @ Four-Leaf — Tue, 13 Oct 2026 16:34:32 +0000

Health insurers in the U.S. process over 5 billion claims a year, according to the Centers for Medicare & Medicaid Services, and a healthcare data analyst spends the job turning that stream into numbers a hospital or health plan can act on. That is why healthcare data analyst interview questions look different from a generic analyst loop. The data under the SQL is coded, regulated and tied to payment.

What do healthcare data analyst interviews actually test?

A healthcare data analyst interview tests whether you can be trusted with data that is regulated, coded and tied to money. SQL fluency gets you into the room, so expect it to be tested. What separates candidates is whether they understand what a row in a claims table or an electronic health record actually represents.

Two kinds of employer hire these analysts, and the interview tilts toward whichever one you are facing. A payer (an insurer or health plan) lives in claims data: one row per billed service line, with a diagnosis code, a procedure code, a date of service and a paid amount that may arrive weeks later. A provider (a hospital system or physician group) lives in clinical and encounter data from its records system, where the questions turn to admissions, length of stay and readmissions.

Read the posting for which side you are on. "Member", "claims" and "utilization" mean payer. "Patient", "encounter" and "EHR" mean provider. Then prepare the domain half of the loop for that side, because the SQL half barely changes.

Which SQL questions come up with claims and EHR data?

One healthcare SQL question you may get is a readmission rate. You get an admissions table and are asked what share of discharges were followed by another admission within 30 days. It tests a self-join on the patient key, a date window, and whether you think to exclude the index admission from matching itself.

The follow-up is where strong candidates separate. The Centers for Medicare & Medicaid Services runs its Hospital Readmissions Reduction Program on 30-day risk-standardized unplanned readmission measures for named conditions and procedures, including heart failure, pneumonia and hip or knee replacement. Your raw rate will not match a CMS figure, and saying why (risk adjustment, planned readmissions removed, a defined condition cohort) shows you know the number has a definition outside your query.

Expect a few other shapes:

Paid versus incurred. Aggregate cost by date of service, then explain why last month's total will rise as late claims arrive.
Duplicates and adjustments. Find claims that appear twice, and tell a true duplicate apart from a reversal and resubmission.
Grain changes. Roll service lines up to a claim, then to a member per month, without double counting.

Say your assumptions out loud before you type. In a healthcare round, an unstated assumption about which date column to use is an easy way for a correct-looking query to answer the wrong question. The SQL interview guide covers the general mechanics of that round.

What should you know about HIPAA before the interview?

Two concepts from the HIPAA Privacy Rule are worth knowing cold, because interviewers use them to test judgment.

The first is minimum necessary. Under 45 CFR 164.502(b), a covered entity must make reasonable efforts to limit protected health information to the minimum necessary to accomplish the intended purpose of a use or disclosure. The same section exempts disclosures to or requests by a health care provider for treatment, which is a useful nuance to have ready. In an interview it shows up as "a manager asks you for a patient-level extract to build a dashboard. What do you send?" The answer starts with the fields the dashboard needs and nothing else.

The second is de-identification. Under 45 CFR 164.514(b), data can be de-identified two ways. One is expert determination, where someone with experience in statistical and scientific principles documents that the re-identification risk is very small. The other is safe harbor, which removes 18 categories of identifiers (lettered A through R in the regulation) and requires that the covered entity has no actual knowledge the remaining data could identify someone. The details trip people up. A ZIP code can keep its first three digits only if that three-digit area contains more than 20,000 people, and ages over 89 have to be grouped into a single category of 90 or older.

Knowing the ZIP and age rules is exactly the kind of specific detail that tells an interviewer you have handled this data before.

Which coding systems will they expect you to read?

A healthcare data analyst is expected to read the code sets in the data and know which system answers which question.

ICD-10-CM is used to code and classify medical diagnoses, per the CDC's National Center for Health Statistics, which maintains it. It answers "what was wrong with the patient." Services and procedures from physicians and other health care professionals are coded in CPT, which CMS describes as Level I of the Healthcare Common Procedure Coding System, maintained by the American Medical Association; CPT codes consist of 5 numeric digits. HCPCS Level II covers products, supplies and services outside CPT, such as ambulance services or durable medical equipment.

The interview question is usually practical. "Find every patient with a diabetes diagnosis" tests whether you match a code family by prefix rather than one exact code, and whether you look at every diagnosis position on the claim instead of only the first one.

Systems exchange this data in standards like HL7 FHIR, which describes itself as a standard for exchanging healthcare information electronically. If the posting names FHIR or an EHR vendor, expect a question about how you would pull data from it, and answer with what you have actually touched.

How do quality measures like HEDIS show up in the interview?

Quality measures are where healthcare analytics meets revenue, and a payer interview may ask you to calculate one. HEDIS, a measure set from the National Committee for Quality Assurance, includes more than 90 measures across 6 domains of care, and NCQA says more than 235 million people are enrolled in plans that report HEDIS results.

Expect a simplified specification and a request to build it: who is in the denominator, what counts toward the numerator, and which exclusions apply. The trap is the order of operations. Apply exclusions to the denominator before you calculate the rate, and say aloud that you are doing so.

On the provider side, readmissions are where the money sits. Under the Hospital Readmissions Reduction Program, CMS caps a hospital's payment reduction at 3 percent, and a candidate who can explain how that program works has an answer ready when the interviewer asks why the rate matters.

What behavioral questions do healthcare analytics teams ask?

Healthcare behavioral questions probe the moments when a wrong number reaches a clinician or a regulator. Prepare two stories in particular.

The first is an error you found after the result had shipped. Interviewers want the sequence. You noticed, you checked, you told the owner before anyone else found it, and you fixed the process that let it through.

The second is explaining a finding to someone who does not trust the data, usually a physician or nurse leader. A strong answer shows you started from their question, used their vocabulary, and showed the limits of the analysis before they found them.

Both work in the structure from our guide to behavioral interview preparation. In healthcare the result lands on care teams and on payment programs, so interviewers listen hardest to how you handled the moment of disclosure.

What is overrated

Long lists of question-and-answer pairs are overrated. Memorizing a scripted answer to "What is HIPAA?" produces exactly the recital that an interviewer follows up past in one question. Memorizing individual ICD-10 or CPT codes is overrated too. The useful knowledge is how code families are structured and where they sit on a claim.

What gets underrated is saying your reasoning out loud while you write the query. The domain knowledge in this post only scores if it arrives in the conversation at the moment the interviewer probes.

A five-step prep plan

Classify the employer. Payer or provider, from the posting's vocabulary. Prepare that side's data model first.
Write the readmission query from scratch, then explain in two sentences why it will not match a CMS figure.
Rehearse the two HIPAA concepts as answers to a scenario. Practice the dashboard extract question until your first sentence is about which fields you would leave out.
Build one quality measure from a published specification, with exclusions, and narrate the order of operations.
Run the loop out loud. Four-Leaf's AI mock interviews include a data analyst track with a recruiter screen, a hiring manager round, SQL live coding and a business case. Name the employer and paste the posting, and the SQL round's questions are built around that posting's domain.

For the statistics and case questions that sit alongside SQL in analytics loops, see how to prepare for a data science interview.

Where this is heading

Healthcare analytics interviews reward judgment about regulated data at least as much as syntax. The harder skill is knowing which fields to leave out of the extract.

Originally published on the Four-Leaf blog.

How to prepare for a machine learning oral exam

Frank @ Four-Leaf — Mon, 12 Oct 2026 03:16:04 +0000

Key takeaways

A machine learning oral exam tests whether you can explain a model out loud and then defend that explanation when the examiner pushes on it. The authors of a 2026 paper on AI-run oral exams at NYU Stern argue that a scripted answer without real understanding breaks down within one or two follow-ups. That makes the useful preparation spoken, done under questioning, and started weeks before the exam rather than the night before.

What happens in a machine learning oral exam?

There's no single format. The three published examples below have one thing in common, which is that you explain something out loud to an examiner who can question you about it.

Task-based, with article skills in the objectives. Aalborg University's 2024/2025 Specialisation Course in Machine Intelligence, part of its MSc in Computer Science (IT), lists an individual oral exam of 45-minutes, with a task handed out 7 days before the examination. The module description says students must present a scientific article's "premises, issue(s), theory, methods, results and conclusions" and assess how practical its proposed solutions are.
Course content plus project. The University of Pisa's Generative and Deep Learning course says its oral examination "will test knowledge of the course contents (models, algorithms and applications)." Students on the traditional route, who don't take the midterms, also deliver a final project and present it with slides for a presentation of 15 minutes during the oral.
Project and case, run by AI. NYU Stern's undergraduate AI/ML Product Management course ran its final oral exam through a voice AI examiner for two cohorts (36 students in Fall 2025, 37 in Spring 2026), as Panos Ipeirotis and Konstantinos Rizakos describe in a paper published in Communications of the ACM. One phase asked about each student's own capstone project. Another walked through a product case study. It's a product management exam rather than a modeling one, but it's the only one of the three with published data on how students experienced it.

Read your own syllabus before anything in this post. The format decides what you practice.

Why do follow-up questions catch students who studied by reading?

Reading builds recognition. You see the bias-variance tradeoff on a slide, it looks familiar, and familiarity feels like knowing it. An oral exam asks for something else, which is producing the explanation from nothing and then extending it in a direction you didn't choose.

The follow-up is the part you can't prepare a script for. In the NYU Stern paper on the AI/ML Product Management exam, the examiner's questions are calibrated to what the student just said, and the authors argue that "a scripted answer that does not reflect actual understanding breaks down within one or two follow-ups." The project phase made this concrete by asking about each student's "goals, data choices, modeling decisions, evaluation approaches, and failure modes," which nobody can improvise convincingly about work they didn't do.

Expect follow-ups from a few families. Why does this work? When does it fail? What would you change if the data looked different? How would you know it was wrong? Each one starts from your previous sentence. That's why a student who can recite the derivation of logistic regression still stalls when asked what happens to it on perfectly separable data.

The practical lesson is that every topic on your list needs two layers. The first is the explanation. The second is the answer to "why" asked twice in a row.

How stressful is an ML oral exam, and does practice fix the nerves?

The best published data says students in one AI-run oral exam found it more stressful than a written one, and offers a small hint that experience didn't help. The safe plan is to prepare to perform while nervous rather than to stop being nervous.

The NYU Stern study surveyed both cohorts of its undergraduate AI/ML Product Management course after their AI-run oral exams. In the first cohort, 83% of respondents found the exam more stressful than written exams. In the second cohort, which sat the exam after several fixes to the examiner, 63% did. In the first cohort, 70% agreed the exam tested their actual understanding. Those numbers come from one course at one institution, so they say nothing reliable about human-run oral exams elsewhere.

One more detail in the NYU Stern data is about experience. You'd expect students who had sat oral exams before to be calmer. They weren't. Every respondent with prior oral-exam experience (all 5 in Fall 2025 and all 4 in Spring 2026) reported the AI exam more stressful. The groups are tiny, so this is a hint rather than a finding, but it's a hint against expecting familiarity alone to calm you down.

What practice does change is how well you think while your heart rate is up. If you've explained gradient descent out loud twenty times, the twenty-first time under stress still has a path to follow. Four-Leaf's guide to interview anxiety covers the in-the-moment techniques, and they transfer directly to an exam room.

What topics should you be able to explain out loud?

Start from the syllabus, then turn each topic into four questions you can answer in under two minutes without notes:

What problem does this solve?
What does it assume about the data?
How is it trained or computed?
How does it fail, and how would you notice?

The fourth question is the one to spend the most time on, because it's where follow-ups naturally go. For a linear model, it's what happens with correlated features. For a decision tree, it's overfitting and what pruning trades away. For a neural network, it's vanishing gradients, overfitting and what each regularizer actually does to the weights.

If your exam centers on a research article, Aalborg University's Machine Intelligence course description makes a good checklist, since it asks students to present an article's premises, problem, theory, methods, results and conclusions. Can you state the paper's premise, the problem it addresses, the theory behind it, its method, its results and its conclusions? Then can you say where the method would break and what the authors didn't test? That last step is what "assess their qualities and practicality" means in practice.

For a project-based exam, rehearse your own decisions. Why that dataset, why that model, why that metric, and what you'd do with another month. Questions about why you built it that way test understanding in a way the finished report can't.

How should you practice for a machine learning oral exam?

Practice the way the exam runs, which means speaking, being interrupted and answering one question at a time.

Explain each topic out loud, cold, and record it. Two minutes, no notes. Play it back and note every place you said "basically" or trailed off. Those are the gaps.
Get asked follow-ups, one at a time. A classmate works well if they ask "why" and "what if" rather than nodding along. In the NYU Stern exam, question stacking was the most common Fall 2025 complaint about the AI examiner, matching a 31% stacked-turn rate, even though its prompt said "You must NEVER ask two questions in the same turn." Your practice partner should follow that rule.
Talk while you write. If your exam has a board or paper, practice writing a derivation while narrating each step. Narrating as you go gives the examiner a chance to redirect you before you're five lines down the wrong path.
Rehearse not knowing. Pick a question you can't answer and practice reasoning toward it out loud from what you do know. A visible line of reasoning gives the examiner something to follow up on, which silence never does.
Run one full mock under time. Use your exam's real length, from your own course's rules, with someone who hasn't seen your notes.

If you don't have a practice partner, a voice mock interview can stand in for the follow-up drill. Four-Leaf's voice mock interviewer is instructed to ask one question per turn and usually follows up on the answer you just gave before moving on. It's designed for job interviews rather than coursework, so bring your own topic list. Four-Leaf's guide to practicing interviews alone covers the recording and mirror methods in more detail.

What is overrated

Rewriting your notes. It feels productive and it builds recognition, which is the skill the exam doesn't test. Time spent tidying a summary sheet is time not spent saying the material out loud.

Memorizing model answers is overrated too, for the reason the NYU Stern paper on AI oral exams gives. A rehearsed paragraph covers the first question and none of the ones after it. Memorize definitions, key equations and the names of things. Then practice saying everything else in your own words, slightly differently each time, so you aren't depending on one phrasing.

How does an ML oral exam compare with an ML engineer interview?

Closely. The authors of the NYU Stern paper on AI oral exams extend their argument past the classroom to "a job interview, a professional licensing board" and anywhere else a polished artifact stands in for understanding that ought to be tested directly. An interviewer asks you to explain regularization, then asks what happens when you double the penalty, then asks how you'd detect that in production.

That overlap is useful if you're finishing a degree and heading into the job market. The habit of explaining a model and then defending it is the same habit the ML fundamentals and system design rounds test. Four-Leaf's breakdown of MLOps and ML engineer interview questions maps the four rounds those loops usually run, and the machine learning engineer interview prep page covers the technical, system design and behavioral questions those loops ask.

The difference is stakes and scope. An exam tests a syllabus you were given. An interview tests whatever the team cares about, plus production questions no course covers. Treat the oral exam as the rehearsal it is.

Where this is heading

The NYU Stern paper on AI oral exams opens with an argument every ML student should take seriously. "Written work no longer certifies that a student understands it," the authors write, and "oral examinations restore that evidentiary link." Their system graded a full cohort's oral exams for under one dollar per exam, on top of a voice subscription that covered the speaking time. When a format that resists ghostwriting gets that cheap to grade, expect to meet it more often.

That makes explaining your work out loud, under questioning, a skill worth building on purpose. Before the exam, get asked "why" twice on every topic, and make sure you have an answer both times.

Originally published on the Four-Leaf blog.

How do recruiters get paid, and what it means for your offer

Frank @ Four-Leaf — Sun, 11 Oct 2026 16:45:40 +0000

The Federal Trade Commission's consumer guidance on job scams puts the business model in one line. "Honest placement firms do not typically charge a fee to job candidates," it says. "Instead, the hiring company pays them a fee to find qualified candidates."

That sentence settles who the customer is. A recruiter who calls you about a role is being paid by the company that has the role, either as a fee per hire or as a wage. Knowing which arrangement you are dealing with tells you what they want from the call, what they will pass along, and how much of your negotiation they can carry.

How do recruiters get paid?

Recruiters who fill permanent roles get paid in one of three ways, and the job title on LinkedIn rarely tells you which.

Contingency agency recruiters are paid only when a candidate they submitted is hired. Top Echelon, which runs a placement network and software for independent recruiters, explains in its guide to agency fee structures that the fee is "typically a percentage of the employee's first-year salary," and that on a contingency search the client "doesn't pay you until someone is hired," however much work the recruiter has done.

Retained search firms are paid part of the fee up front to run an exclusive search. The same Top Echelon guide says the total fee percentage on a retained search "is typically greater than the fee percentage" on a contingency agreement, because the searches run longer.

In-house recruiters are employees of the company that is hiring. The U.S. Bureau of Labor Statistics counts them inside human resources specialists, noting that recruitment specialists are "sometimes known as recruiters or talent acquisition specialists." The median annual wage for that occupation was $75,940 in May 2025, per the BLS Occupational Outlook Handbook.

Staffing agencies, which employ temporary and contract workers, are a separate business and get their own section below.

How much does a company pay a recruiting agency?

A company typically pays a contingency agency about a fifth of the hire's first-year salary. The cleanest published numbers come from Top Echelon's own network. For placements its members made in 2018, the average fee percentage was 21.5%, the average fee was $20,283 and the average starting salary was $93,407. Information technology placements in the same data averaged a 22% fee, or $21,956 per hire.

Those figures are old and come from one network of independent recruiters, so read them as an order of magnitude rather than a current market rate. Top Echelon's 2021 guide to agency fee structures also notes that some agencies charge flat fees instead of a percentage.

The order of magnitude is the useful part. Benchmarking data from the Society for Human Resource Management, reported in April 2022, put the average cost per hire at nearly $4,700. The samples and years differ, so the two numbers do not divide cleanly. They still suggest that an agency hire costs an employer several times what a typical hire costs. The same candidate can therefore carry a different price depending on how they arrive. A referral or a direct application carries no agency fee. An agency submission carries one for as long as the agency's claim on you lasts.

Does an agency recruiter want you to get a higher salary?

The fee structure pulls an agency recruiter two ways. Top Echelon's 2021 guide to agency fee structures says the fee is a percentage of the hire's first-year salary, and that on a contingency search the client pays nothing until someone is hired. A higher offer means a higher fee. An offer that falls apart means no fee. Which pressure wins depends on the recruiter and the deal, so judge an agency recruiter by what they do when you push.

Guarantee clauses add a third pressure. Top Echelon's guide to collecting placement fees lists the "guarantee period" among the terms a recruiter should settle with a client. The published agency terms of Mallesons, a law firm that sets out the rules agents must follow for its Australia and Singapore offices, require the agent to refund or credit its fee if the hire does not stay for "the duration of any probation period." An agency on terms like that has to refund or credit its fee if you take a role that fits badly and leave early, which gives it a reason to care about fit.

The Federal Trade Commission's job-scam guidance describes honest placement firms as paid by the hiring company, so treat an agency recruiter as a broker the employer pays. Tell them what you need in order to accept.

How are in-house recruiters paid, and whose side are they on?

In-house recruiters are employees of the company that is hiring, and they speak for it. BLS figures put the median annual wage for human resources specialists, the occupation that includes recruiters, at $75,940 in May 2025. That figure says nothing about whether a bonus rides on your offer, so do not assume anything in either direction.

When an in-house recruiter asks what you are looking for early in the process, they are asking on the employer's behalf. Four-Leaf's guide to the recruiter screen covers how to handle that question without giving away your number, and the post on what hiring managers do when you negotiate walks through what happens to a counter once you send it.

An in-house recruiter is also someone you may work alongside if you join. Keep the negotiation specific and polite, ask them how the process runs and when decisions get made, and treat their answers about timing as the employer's answers.

What happens when a recruiter submits your resume?

When an agency submits your resume, the contract between that agency and the employer decides what the submission means, and candidates rarely see it.

Mallesons publishes its version. Its terms say an agent "must not submit a candidate" without "express written permission from the candidate" for that specific role. An agent with that permission then owns the candidate for that role for twelve months. If two agents claim the same candidate, they have to settle it between themselves, and while that dispute is open "Mallesons is not required to pay any Fee."

Those are one firm's terms, and other employers write their own. They still show what is worth asking about. Under the Mallesons terms an agency needs your written permission for each role, and two agencies claiming you for the same role leave the firm free to withhold the fee while the dispute is open. Ask every agency recruiter which companies and roles they plan to send your resume to, and to wait for your written yes on each one.

What about staffing agencies and contract roles?

Staffing agencies employ temporary and contract workers and place them with client companies, which makes them a separate business from permanent placement. It is a large one. The American Staffing Association reports that nearly 2.2 million temporary and contract employees worked for U.S. staffing companies in an average week in 2024. The association also reports that 64% of staffing employees work in the industry to fill the gap between jobs or to help them land a job.

For anyone using contract work as a bridge, the clause to ask about is conversion. Top Echelon's 2021 guide to agency fee structures describes how an agency can charge a client that wants to hire a contractor permanently, one option being the usual placement fee prorated against how long the contractor has already worked. A client facing that fee may wait for it to shrink before converting you. Asking the agency how conversion works before you start tells you what the "temp-to-perm" in the listing actually means.

What is overrated

Advice that treats every recruiter the same way is overrated, whether it casts them as your advocate or as an obstacle. The payment model is a better guide than either label. A contingency agency is paid when a candidate it submitted is hired, a retained search firm is paid a retainer up front and further payments later, per Top Echelon's 2021 guide to agency fee structures, and an in-house recruiter draws a wage from the employer. Three payment models give three people different reasons to want the same offer signed, and different limits on what they will push for.

Five steps for working with recruiters

Ask how they are paid in the first call. "Are you in-house or with an agency, and is this retained or contingency?" is a normal question, and the answer frames everything after it.
Give permission per company, in writing. Never let an agency send your resume anywhere you have not approved by name.
Keep a list of where each agency has sent you. If two recruiters offer to submit you for the same role, pick one, because under terms like the Mallesons agency terms a double claim leaves the employer free to withhold the fee while the dispute is open.
Share your requirements, and keep your history. Tell recruiters what you need to accept. Your current pay and your other offers are leverage you spend once.
Rehearse the money conversation out loud. Four-Leaf's salary negotiation coach runs voice practice scenarios in which an AI recruiter asks for your number first, or holds base salary at the top of the band and waits to see whether you ask about anything else, so the first time you hear those lines is not the real call.

Where this is heading

A recruiter conversation is easier to read once you know who pays for it, how and when. That fact matters more than the recruiter's title or tone, and it takes one question to find out. The candidates who handle recruiters best know which kind is on the line, and say what they need without saying what they would settle for.

Originally published on the Four-Leaf blog.

The Atlassian interview process, from Atlassian's own handbooks

Frank @ Four-Leaf — Fri, 09 Oct 2026 16:45:15 +0000

Key takeaways

Atlassian publishes a separate interview handbook for engineering, product and design candidates, and all three describe the same round. Every candidate sits a values interview, likely with someone from outside the hiring team, and for engineering roles the final call goes to a Hiring Committee that works from interview feedback and your CV. Both facts are worth preparing for as deliberately as the coding rounds in the Atlassian interview process.

What are the stages of the Atlassian engineering interview?

Atlassian's engineering interview handbook walks through four interviews and a final review.

Coding. The coding interview comes in two parts, data structures and code design. Atlassian says you can choose the language you work in, because it assesses problem-solving and learning agility rather than proficiency in one stack. The handbook is explicit that missing a line won't cost you the job, and that it wants to see how you weigh trade-offs like verbosity against optimization.

Systems design. A 60-minute technical interview where the goal is not code. The interviewer picks from a set of structured questions about the kinds of problems Atlassian solves, and wants to hear what questions you would ask, how you would talk through reliability and cost, and which technologies you would use.

Manager interview. Usually one-on-one with the hiring manager or a more senior manager, covering where you want your career to go and how you work. Atlassian asks you to be ready to talk through a past project or two, and says you could also speak to the business justification for it.

Values interview. A separate interview on Atlassian's five values, likely with someone from outside the hiring team.

Hiring Committee. After the interviewers debrief, a good fit moves to the committee for the final decision.

What is the Atlassian values interview?

The values interview assesses how well you align with Atlassian's five values, and the engineering handbook says your values interviewer likely won't be a member of the team you're applying to join. It could be someone from Sales, HR, or Customer Support. Atlassian's design interview handbook puts it at 45 minutes, calls the sessions relaxed and conversational, and says questions are tailored for managers and individual contributors.

The design handbook also gives the clearest instruction on format. Atlassian encourages the STAR method for this interview, and says you can draw on any experience, including but not limited to work.

The values themselves are listed on Atlassian's values page.

Open company, no bullshit
Build with heart and balance
Don't #@!% the customer
Play, as a team
Be the change you seek

The practical consequence of an outside interviewer is that technical jargon is unlikely to help in this room. A support lead hearing your story about pushing back on a launch can follow the customer reasoning far more easily than the architecture. Prepare one story per value, and tell each one so that a person from a different department can follow it.

Who makes the final hiring decision at Atlassian?

For engineering roles, the engineering handbook says a Hiring Committee makes the final call. Atlassian introduced it "to ensure a consistent and fair interview experience," and the committee works independently from the interviewers. It only has access to specific information such as interview feedback and CV information, and it looks holistically at your skills, strengths and behaviors.

That changes what a good interview is. The committee works from interview feedback and your CV, so whatever you do well has to survive being summarized by someone else. Name the trade-off you chose and the reason in one plain sentence, because that sentence is the one an interviewer can carry into their feedback.

The other two handbooks describe a different last step. Atlassian's product interview handbook says all 4 interviewers meet to discuss their hiring recommendation with the hiring manager, and to work out why if they are not aligned. The design handbook describes a calibration session between all interviewers. If you are interviewing outside engineering, ask your recruiter which process applies to your role rather than assuming the engineering one.

How do Atlassian's product and design interviews differ?

The product track uses a panel after the first hiring manager conversation. Atlassian's product handbook says that panel is 3 interviews covering its expectations for product managers, "one for each pillar," plus 1 values interview. The handbook lists four pillars (lead and inspire, craft mastery, deliver outcomes, and great communicator) and says communication is assessed throughout, so the three panel interviews most likely map to the other three. The handbook asks for outcomes over outputs, meaning what a project moved for customers or the business rather than how long it took, and says interviewers hope to hear candidates using "we" more than "I." Our product manager interview questions guide covers the question types.

The design track starts with a 30-minute recruiter screen and a 60-minute hiring manager interview, which includes a 10-minute case study and 10 minutes of Q&A about it. A 45-minute portfolio review follows, then a 60-minute portfolio follow-up with two interviewers covering product thinking and craft excellence. Senior individual contributors and managers may also meet engineering and product counterparts in what Atlassian calls a triad interview. The values interview closes the loop.

What does Atlassian ask of early career candidates?

Atlassian's early careers interview guide adds a gate before any interview. Graduates and interns take an initial assessment specific to the role, and the guide says it is mandatory to clear it to reach the interview process. Its examples are an online coding test for a software engineering graduate role and a take-home case study for a product management graduate role, which in most cases must be finished within a set timeframe.

After the assessment, the guide says to be prepared for 3 to 4 interviews, which may include craft-specific technical and functional assessments plus interviews on leadership and values. Atlassian runs its intern and graduate interviews virtual-first. The guide recommends the S.T.A.R. method for behavioral questions and suggests preparing an elevator pitch of under two minutes for the "tell me about yourself" question.

What is overrated in Atlassian interview prep?

Grinding company-tagged coding problems. Atlassian's engineering handbook says the coding round draws from a bank of questions, lets you pick the language, and cares most about how you think. The handbook says its interview questions are designed to "ladder," with follow-ups that get harder or easier as you go, which in systems design is a test of how you adjust when the interviewer adds a trade-off. Neither rewards a memorized solution. Our system design interview questions guide is more useful preparation than a list of leaked prompts.

Underrated, by contrast, is the values round, because it sounds informal. Atlassian's design handbook says every candidate has a values interview, a full 45-minute slot, likely with someone who is not on the team you would join.

How to prepare for an Atlassian interview

Read the handbook for your track before the first call. Atlassian publishes separate ones for engineering, product, design and early careers, and the stages differ.
Pick your coding language now and practice in it. Atlassian lets you choose, so choose the one you think in fastest.
For systems design, practice saying your constraints out loud before you draw anything. Reliability and cost are the two the handbook names.
Write five values stories, one per value, each in STAR shape with a result a non-engineer can follow.
Rehearse those stories out loud. A story that reads well on paper often runs long or loses its result when spoken, and a voice mock interview is one way to hear that before an outside interviewer does.
Prepare questions for the values interviewer about how the values show up day to day. Atlassian's design handbook says every interview should be a two-way conversation.

The pattern worth taking away

Atlassian's handbooks spell out two things worth planning around. They tell you your values interviewer is likely to come from another part of the company, and for engineering roles they tell you the decision belongs to a committee working from information such as interview feedback and your CV.

Both point the same way. Prepare answers that still make sense when someone else repeats them.

Originally published on the Four-Leaf blog.

The Adyen interview process, in Adyen's own words

Frank @ Four-Leaf — Thu, 08 Oct 2026 16:49:35 +0000

Key takeaways

Adyen, the Amsterdam-founded payments company, publishes its hiring process in detail. Six stages, a hiring philosophy built on eight written principles, and a last round that puts every candidate in front of a senior executive from its Management Board or Global Leadership team. Its recruiters have also written down, by name, what they want to hear. The short version is that Adyen tests whether you know the company and whether you can tell specific, honest stories about your own work. Both are things you can prepare for.

What are the stages of the Adyen interview process?

Adyen's How we hire page names six stages and says they "apply across all teams and regions," while warning that the order may vary slightly.

Application review. Adyen commits to being in touch within five business days of receiving your application. That gives you a useful marker. If a week passes in silence, a polite check-in is reasonable.

Recruiter screen. A recruiter reaches out to learn about you and the experience relevant to the role. Treat it as the first real interview. Separately, Adyen's recruitment team tells candidates to use "recruiter prep calls" to ask what to expect, so the recruiter is also your best source on the rounds that follow.

Team interview. You meet someone from the hiring team for an in-depth conversation about the role, your experience, and how you work.

Skills assessment. Depending on the role, Adyen asks for "either a technical skills test or a case study presentation." More on this below.

Leadership interview. You meet a leader from the team you're joining and discuss your motivation for joining Adyen.

Final interview. Per Adyen, your final interview "will be with a senior executive from our Management Board or Global Leadership team."

One variation is worth knowing in advance. In a September 2025 post from its Global Recruitment team, Adyen notes that "for some roles, the order of step three and five will be reversed," meaning the leadership conversation can come before the team interview. Ask your recruiter which order you're on, since it changes who you need your motivation story ready for first.

What is the Adyen Formula, and why does it come up in every round?

Adyen says its hiring process is "rooted in the Adyen Formula," and the Formula page lists the eight principles that guide how the company works. A few of them read like interview prompts already. "We launch fast and iterate." "Winning is more important than ego." "We don't hide behind email," with the rest of the line being that Adyen staff pick up the phone instead. "We talk straight without being rude."

Plenty of companies publish values. What makes Adyen's worth taking seriously is that its recruiters have said what they want candidates to do with them. The Global Recruitment team writes that it wants "engagement with the Formula past pure memorization," and that candidates should find connections with their past experiences. A São Paulo-based recruitment partner quoted in the same post encourages candidates to ask interviewers about the Formula from their own standpoint, adding that "everyone will be 100% glad to talk about it."

Read that as a warning against reciting the list. Pick the principles that genuinely describe how you work, and find the moments in your career that prove it. A time you delivered unwelcome news to a senior colleague, clearly and without heat, says more about talking straight without being rude than any paraphrase of the line. And if a principle sits badly with how you like to work, the interview process is a good time to find that out.

What happens in the Adyen skills assessment?

Adyen publishes little about the format beyond the split, which makes sense for a company whose careers site spans commercial, tech, product, and operational roles. Depending on the role, you get either a technical skills test or a case study you present, and your recruiter can tell you which.

What Adyen does say is how it judges the result. Deborah Desouza, an Amsterdam-based recruiting lead quoted in the recruitment team's post, says that "we really want to see how you think on your feet, and preparation goes a long way." She adds that the assessment simulates "a real-world scenario with real-world problems," and that candidates should articulate how they'd think through the solutions.

Read those two lines together and the instruction is to narrate. A correct answer delivered in silence gives the interviewer very little to score. A slightly rougher answer where you state your assumptions, explain the tradeoff you're making, and say what you'd check next shows them the thinking they asked to see. For a case presentation, that means rehearsing the delivery as well as building the slides. For a technical test, it means practicing talking while you work, which feels unnatural until you've done it a few times. Our take-home assignment guide covers how to scope and present work like this.

Who do you meet in the Adyen final interview?

Every candidate's last round is with a senior executive from the Management Board or Global Leadership team, according to Adyen's own hiring page. For a company whose about page lists 4000+ employees and 115+ nationalities, putting a senior leader in front of each hire is a real cost. Adyen doesn't publish what that round assesses, so ask your recruiter.

If you're interviewing for a leadership position, Adyen has said what it wants to learn across the process. Matt Rum, a San Francisco-based executive and leadership recruiter, says the team wants to explore "how you think, how you've grown, and how you've guided others through complexity." Whatever your level, the leadership interview before this round is explicitly about your motivation for joining Adyen, so know why you want Adyen in particular.

How long does the Adyen interview process take?

Adyen's FAQ says the process takes "a little over a month" on average and varies by role. If you have a constraint, such as a competing offer or a notice period, the same page asks you to tell your recruiter so they can try to accommodate it.

A few other practical details from that FAQ are worth knowing. Interviews happen by phone, via Zoom, and in person. There is no dress code. If you're declined for a role, there is a 30-day waiting period before you can reapply for that same position, and Adyen's applicant tracking system enforces it. And to help candidates spot impersonators, Adyen states that legitimate messages come only from addresses ending in @adyen.com and that it never asks for payment or financial information.

Where engineering-focused guides fall short

The prep-site guide we read before writing this one is built for engineers, around Java coding rounds and system design, with the Formula as one round among five or six. Adyen's own page puts it differently. The company says it expects candidates to have "a solid understanding of our business, our products, and the Adyen Formula," on top of the technical or role-specific requirements.

A candidate who has drilled algorithms but can't explain what Adyen sells, who its customers are, or which principle a past project illustrates is underprepared by Adyen's own published standard. The recruitment team even names the reading list. It points candidates to the job description, the Formula, customer case studies, and the company's latest earnings report.

How to prepare for an Adyen interview

Read Adyen's latest earnings report and two customer case studies before the recruiter screen. Be ready to say in two sentences what Adyen does and why that business interests you.
Read the Formula page and pick the principles that honestly describe how you work, with a real example behind each. Use the STAR method, which Adyen's recruitment team recommends by name, to keep each one tight.
Book the recruiter prep call and ask three things. Which order your stages run in, what format your skills assessment takes, and who you'll meet in the final round.
Rehearse your skills assessment out loud, narrating your reasoning as you go. Adyen says it wants to see how you think on your feet, and that only shows if you speak while you think.
Practice your stories out loud under time pressure before the team interview. Ashley Gomez, a Singapore-based Senior Recruitment Business Partner at Adyen, tells candidates to "be honest, lead with specifics, and be a storyteller," and the gap between knowing your answer and delivering it under pressure is where most specifics get lost. A voice mock interview is a quick way to hear whether yours survive.
Prepare real questions for each interviewer, including one about how they apply the Formula day to day. Our list of questions to ask at the end of an interview is a starting point.

The pattern worth taking away

Adyen has written down what it wants from candidates in plain terms. Know the business. Connect its principles to things you've actually done. Show your thinking in the skills assessment. Be ready to talk about who you are with a senior executive. None of that rewards cramming the night before, and all of it rewards a candidate who did the reading and practiced telling their own stories out loud.

The companies that publish their process this openly are telling you the test in advance. The candidates who win are the ones who take them at their word.

Originally published on the Four-Leaf blog.

The IBM interview process, from IBM's own careers pages

Frank @ Four-Leaf — Thu, 08 Oct 2026 16:48:42 +0000

Key takeaways

IBM says more than three million job seekers apply to it each year, and its careers site spells out what happens to them. Five steps, up to two online assessments IBM might ask for before the interview, structured behavioral interviews, and a written policy on exactly where AI is and isn't allowed. Almost every stage of the IBM interview process comes with a published rule you can prepare for in advance.

What are the steps in the IBM hiring process?

IBM's application process page lists five steps, with a caveat that the hiring process may vary by location and region.

Application. You apply online, one job at a time, and track status in IBM's candidate portal. The FAQs on that page say applications cannot be edited after submission. If you made a mistake, the fix is to withdraw and reapply while the role is still open, so check the visa status and start date fields before you submit.

Screening. In IBM's words, "IBMers with expertise in your field will review your application, skills and experience." If your status flips to "No Longer Under Consideration" right after you apply, IBM says it usually means your application did not meet one or more basic job requirements. That points at a hard requirement in the posting, so reread it before applying elsewhere.

Online assessment. IBM might ask for one or two assessments, which vary by role. More on these below.

Interview. Candidates who successfully complete the assessments may be invited to a phone, video or in-person interview, or to an assessment center visit, which IBM describes as a group exercise and an in-person interview.

Decision. IBM shares its decision after your final interview.

Which online assessments does IBM use?

IBM's application page says "we might ask you to complete one or two assessments, which can vary depending on your role," and describes three of them. A coding assessment of multiple-choice questions or coding challenges. A video assessment, where you answer a series of interview questions using your webcam and microphone. An English language assessment of multiple-choice questions in a text format. IBM's separate assessment FAQ adds a fourth type, IBM work behaviours assessments.

The assessment FAQ carries the practical advice. IBM asks candidates to complete their assessments within 7 days of receiving the invite, to check whether each one is timed, and to set aside enough time to finish in one sitting. On the work behaviours format, it tells you that being asked to pick between two statements is normal, and to "choose the statement that you agree with slightly more." The same page advises going with your first reaction rather than overthinking.

The video assessment is worth rehearsing on its own, because there is no person on the other side to react to. If you have never recorded answers to a camera, our HireVue practice guide covers how one-way video interviews work and how to rehearse for them. Disability accommodations are available, and IBM says candidates invited to an assessment receive instructions on how to request one.

What is IBM's policy on using AI during hiring?

IBM publishes an "Acceptable Use of AI by Candidates During Recruitment" section on its application page, and it is specific. Permitted uses include improving the grammar and clarity of a resume, understanding a job description, researching common questions, and "Practicing role play mock interviews." IBM encourages candidates to disclose significant AI assistance in submitted materials, while minor grammar and formatting edits need no disclosure.

The prohibited list is just as clear. No AI to generate an entire resume or cover letter without personal input. No AI to complete assessments, technical tests or coding challenges. No AI "during live interviews or real-time assessments." IBM's own scenario table gives the example of having ChatGPT open during a Zoom interview to suggest answers in real time, and marks it not allowed. The page warns that a violation may result in disqualification from the recruitment process.

Read the two lists together and AI is welcome as a practice partner before the interview and unwelcome as a hidden prompter during it. On the live interview, that matches our own position on real-time answer tools. If you are weighing a real-time answer tool for any employer, our piece on why we don't build interview copilots lays out the risks.

What kind of interviews does IBM run?

The FAQs on IBM's application process page say it uses "what we call structured or behavioral interviews," with questions about the technical skills and behaviors the role requires. Its advice for preparing is to think of specific examples of when you have proven those skills, and then to focus on outcomes. What did you achieve? How did you measure your success? What role did you play in making it happen? IBM adds that interviewers might ask about your short-, mid-, and long-term career goals.

For technical roles, the same FAQs point candidates to LeetCode, Pramp and HackerRank to take as many practice tests as possible, covering both the coding assessment and the technical interview.

IBMers quoted on IBM's own careers blog are more concrete still. In a post of interview tips from IBM employees, a Q2C practice leader named Raymond advises candidates to "storyboard 3 to 4 stories that demonstrate skills required for the position," covering your role, your actions, and the results and metrics. A business analyst named Thayna sums up the approach as "Think, Prepare, Rehearse." The STAR method is the standard shape for stories like these, and IBM's emphasis on measured outcomes means the result step deserves the most work.

How long does the IBM hiring process take?

IBM declines to put a number on it. The FAQs on its application process page say the length depends on the role, how quickly the applicant responds to follow-up questions, and how many applications recruiters receive, "so there is no overall timeframe we can share."

What you can do is watch the candidate portal, where IBM says your status updates as your application moves and you also receive emails when it changes. "Under Recruiter Review" means a recruiter is actively looking at your application and no decision has been made. IBM also says that, given its application volume, it is not always possible for recruiters to contact applicants directly, so the portal and its status emails are the signal to watch.

One safety note from the same page. IBM says all legitimate correspondence from it will originate from an ibm.com email address, and that it does not use personal email domains like gmail.com for recruiting. An "IBM recruiter" writing from a free email account is a scam signal.

What IBM's pages leave for you to find out

IBM's process is described at the level of the whole company, across every kind of role it hires for. That means some of the most useful details for your specific role are left to the recruiter. Which assessments you will get and the specific questions you face vary by role, and the process as a whole may vary by location and region. Whether your interview is a single conversation or an assessment center day is not something the pages can tell you in advance.

Treat that as an invitation to ask. A separate IBM careers blog post of advice from IBM recruiters tells candidates to ask their own questions "to find out more about the role, manager, selection process, and company itself." A candidate who asks early can find out which format to rehearse for. After an interview, Sabhyasachi in India, one of the talent acquisition colleagues quoted there, suggests asking your recruiter for feedback "to help you identify areas of improvement."

How to prepare for an IBM interview

Reread the posting's basic requirements before you apply, and check every field before you submit, because IBM applications cannot be edited afterward.
Block time for the assessments as soon as the invite lands. IBM asks for completion within 7 days, and some assessments are timed and best done in one sitting.
Practice answering questions to a camera if your role includes the video assessment. Record yourself and watch it back once.
Write three or four stories that each prove a skill from the job description, with a measured result in every one. The FAQs on IBM's application process page ask what you achieved and how you measured it, so a story without a number or a concrete outcome is unfinished.
Rehearse those stories out loud before the interview. Thinking through an answer and delivering it under pressure are different skills, and IBM's own policy names role-play mock interviews as an acceptable use of AI. A voice mock interview is one way to hear whether your outcomes survive the telling.
Prepare questions about the team and the selection process, and ask your recruiter what format comes next.

The pattern worth taking away

IBM has published a large share of its rulebook. Up to two assessments, a request to finish them within a week, a structured behavioral interview built on outcomes, and a plain statement of what AI use may get you disqualified. Most of it can be prepared for before the first invite arrives.

The preparation that follows from those pages is simple to state. Practice with every tool IBM allows, then walk into the live round with nothing but your own stories.

Originally published on the Four-Leaf blog.

Project manager job outlook for 2026, read from the BLS data

Frank @ Four-Leaf — Mon, 05 Oct 2026 16:47:54 +0000

The U.S. Bureau of Labor Statistics projects employment of project management specialists to grow 7 percent from 2025 to 2035, against 3% for all occupations. It also projects about 76,500 openings for the role every year of that decade. Set against the 73,200 jobs that growth adds over the whole decade, those openings show where most of the hiring comes from. Most of them replace people who leave the occupation, and that changes how a candidate should search, apply, and interview.

What is the project manager job outlook for 2025 to 2035?

The Bureau of Labor Statistics Occupational Outlook Handbook entry for project management specialists, last updated in August 2026, is the most direct federal answer. It counts 1,094,300 jobs in the occupation in 2025 and projects 1,167,400 by 2035, a gain of 73,200. BLS labels the 7 percent growth rate "much faster than the average for all occupations."

The reasons BLS gives are ordinary ones. Organizations want to maintain and improve productivity, projects need to finish on time and within budget, and information technology projects keep growing in volume and complexity. None of that depends on a hiring boom.

O*NET, the Department of Labor's occupational database, also tags project management specialists as a "Bright Outlook" occupation. Its sample of reported job titles shows how wide the label runs, from "Implementation Project Manager" to "Fiber Project Manager."

The reading that fits the data is steady demand. BLS projects faster than average growth plus tens of thousands of replacement openings a year, which describes an occupation where employers hire every year.

Where do 76,500 project manager openings a year come from?

The Bureau of Labor Statistics projects about 76,500 openings a year for project management specialists, but employment growth accounts for only 73,200 new jobs across the whole ten years. The rest of the openings exist because people leave. BLS says many of them come "from the need to replace workers who transfer to different occupations or exit the labor force, such as to retire."

That has two practical consequences for a project manager looking for work.

First, a replacement opening appears when a specific person moves on, so the projection gives no reason to wait for a single hiring window. A search that runs steadily is likely to see more of those openings than one that stops and starts.

Second, when a role is a replacement, the hiring manager may already know what the last project manager did well and what fell through the cracks. That history can shape what the interview probes, which is why asking "what did the previous person in this role struggle with?" is one of the most useful questions a candidate can bring.

Is "project manager" one job or three in the federal data?

The title "project manager" is spread across at least three occupations in the federal data, and only two of them come with their own BLS projection.

Occupation	Jobs in 2025	Projected growth, 2025 to 2035	Median pay, May 2025
Project management specialists (BLS)	1,094,300	7 percent	$102,320
Construction managers (BLS)	609,100	9 percent	$114,990
Information technology project managers (O*NET)	Not published separately	Not published separately	Not published separately

The Bureau of Labor Statistics says outright that construction managers are "often called general contractors or project managers," so a construction project manager reading the project management specialist figures is reading the wrong row.

IT project managers have their own O*NET profile, Information Technology Project Managers, which O*NET also tags as a Bright Outlook occupation. Its wage and employment data, though, is reported for "Computer Occupations, All Other," a catch-all group. No federal source publishes a growth rate or a median that belongs to IT project managers alone, so any article quoting one is quoting something else.

The practical point is that a candidate quoting one growth number is quoting one slice. A construction project manager and a software delivery manager share a title and almost nothing else in how they are hired, what they are paid, or what the interview tests.

How much do project managers earn, and what moves the number?

The Bureau of Labor Statistics puts the median annual wage for project management specialists at $102,320 in May 2025. The spread is wide. The lowest 10 percent earned less than $61,580, and the highest 10 percent earned more than $167,970.

Industry moves the median, though it accounts for only part of that range. Among the five top industries BLS lists for project management specialists, finance and insurance paid the highest median, at $120,700, and construction paid the lowest. Construction project managers counted under the separate construction managers occupation are a different case, with a median of $114,990 for that occupation as a whole.

Two things follow for a job seeker. A project manager moving between industries is changing pay bands as much as changing employers, so a salary ask should be benchmarked against the target industry rather than the current one. And because the 10th and 90th percentiles sit so far apart, a single "average project manager salary" figure from a listicle is close to useless in a negotiation. Four-Leaf's guide on negotiating salary without leverage covers how to anchor an ask when you have no competing offer.

Which industries hire the most project managers?

Professional, scientific, and technical services is the largest employer of project management specialists, holding 28% of the occupation's jobs in 2025, according to the Bureau of Labor Statistics. Construction is second. Manufacturing, administrative and support services, and finance and insurance follow well behind.

Construction in second place is the useful detail for anyone who pictures project management as a software job with sprints, standups and Jira boards. The Handbook does not break down the professional services share, which can include consulting, engineering, and computer systems design work, but construction is a large slice of the occupation on its own.

A search that runs only on tech job boards sees a fraction of the market. A candidate with transferable skills should search by the work (scheduling, budgets, vendor management, stakeholder reporting) as well as by the title.

What do project manager interviews actually test?

Project manager interviews test whether a hiring manager can trust a candidate to deliver when the plan breaks. The Bureau of Labor Statistics lists the qualities it considers important for the job, among them communication, problem-solving, and time management under tight deadlines, and interview questions are a practical way to check those under pressure.

Expect three families of question, which Four-Leaf's project manager interview prep page breaks down in more detail. Scenario questions ask what you do when a critical vendor slips and the date cannot move. Behavioral questions ask about leading without authority, managing up when priorities change, and a project that missed its deadline. Technical questions cover fundamentals like critical path, Agile and Waterfall trade-offs, and the tools the team uses.

The behavioral family deserves the most rehearsal, because every candidate claims to be organized and only a specific story shows it. The candidates who stand out describe a real failure in plain terms and say what they changed afterward. Four-Leaf's guide to behavioral interviews covers how to build those stories.

What is overrated

The headline growth rate is overrated as a reason to choose or leave the field. A 7 percent rise over ten years is real, but the 76,500 annual openings matter far more to anyone applying this year, and those come mostly from replacing people who leave.

Certification as a gate is overrated too. The Bureau of Labor Statistics says certification is "not always required" for project management specialists, although it can demonstrate competency to employers. Some postings do require one, so read them. Where a posting does not require one, concrete project results are what an interviewer can probe. BLS notes that some positions require relevant work experience and that candidates may gain it in roles such as business analyst.

Generic salary averages are the third. The Bureau of Labor Statistics figures put the line where the top tenth of project management specialists starts at more than two and a half times the line where the bottom tenth ends, which makes any single figure a poor anchor.

A plan for this week

Decide which of the three occupations you are actually targeting. Use the BLS pay and growth figures for project management specialists or construction managers if one of those is your target, and treat any single growth number quoted for IT project managers with suspicion, since no federal source publishes one.
Search by the work as well as the title. Add "program coordinator," "implementation manager," and "delivery lead" to your saved searches, and include construction and professional services employers.
Rewrite three resume bullets so each one names a budget, a timeline, or a team size you actually managed.
Prepare one story about a project that slipped, told honestly, with the specific change you made afterward.
Say your answers out loud. Practice a vendor-delay scenario and a leading-without-authority story until each lands in under two minutes. Four-Leaf's voice mock interviews run this kind of rehearsal and score the spoken answer.

Where project management hiring is heading

The Bureau of Labor Statistics projection describes a field that keeps hiring because people keep moving through it, with growth on top. That is a good market for a prepared candidate, because the openings keep appearing one replacement at a time.

The candidates who win these roles are the ones who can prove what they delivered. A project manager's whole job is making a plan survive contact with reality. The interview is the first project a hiring manager gets to watch you run.

Originally published on the Four-Leaf blog.

The Snap Inc interview process, from S.A.I.L. to AI-assisted coding

Frank @ Four-Leaf — Sat, 03 Oct 2026 16:45:09 +0000

Key takeaways

Snap Inc asks candidates to answer behavioral questions in a format it calls S.A.I.L., short for Situation, Action, Impact, Learning. It overlaps with the more familiar STAR method and adds a fourth step. Its How We Interview page also says AI is not allowed in interviews unless you are specifically told otherwise, while its technical hiring page says your AI fluency may be evaluated. Both are true at once, and the software engineering loop shows how. Each of the four tracks on Snap's software engineering interview prep page lists an AI Assisted Coding round.

What are the stages of the Snap interview process?

Snap's technical and product hiring page lays out five steps, and it says up front that your specific steps might vary because the teams are as varied as the platforms.

Apply. Through Snap's careers site.

Recruiter screen. The first conversation, where Snap learns your background and you learn about the role and the process.

Initial interview. Snap's technical hiring page says this round assesses a values-based behavioral question alongside your skills for the role. Depending on the role, it may include coding fundamentals, domain knowledge, or both. Expect to talk through specific examples of your work, including metrics and leadership.

Virtual loop. A series of interviews tailored to your role and level, typically mixing values-based behavioral questions with technical areas for the job. Snap's technical hiring page says that for a virtual onsite, Snap will do its best to accommodate a split over two days. Technical portions run on HackerRank.

Offer. Some roles are tied to a specific team from the start. Others include team matching toward the end, where Snap's page says your experience, interests and location preferences are taken into account.

The values question shows up early. It appears in the initial interview, before the loop, so a weak behavioral story could cost you before you ever reach a system design question.

What is the S.A.I.L. method Snap uses?

Snap's How We Interview page says that when you answer behavioral questions, Snap is interested in four details, and it spells out what each one means.

Situation. What's the background, and what problem needed to be solved?
Action. What was your role, and what actions did you take?
Impact. What were the results, and what impact did it have?
Learning. What did you take away from the experience?

The first three steps cover the same ground as STAR, with Impact in place of Result. The fourth is the one a STAR-trained candidate is least likely to have rehearsed. A STAR answer ends when the result lands. A S.A.I.L. answer keeps going for one more beat and says what changed in how you work.

That last beat is easy to fake and easy to spot. "I learned the value of communication" is a sentence an interviewer has heard a hundred times. A strong Learning step names a specific behavior you changed afterward and, ideally, a later situation where you used it. Draft your stories with all four steps written out, then cut the Situation down. Spend too long on setup and you can run out of time before the Learning step.

What does Snap mean by Craft and Value competencies?

Snap's How We Interview page describes its approach as competency-based interviewing, a structured method that assesses your past experiences, behaviors and skills against the specific requirements of the role. Snap calls those requirements Craft & Value Competencies.

The page draws a clear line between the two. Values-based competencies are always assessed with behavioral questions. Craft competencies are assessed with either behavioral questions or demonstrations of skill, which for an engineer likely means the coding and design rounds.

Snap's How We Interview page groups its values in action under Kind, Smart and Creative, says they are woven into the interview process, and lists these traits with a one-line definition for each. They are Courage, Action Oriented, Manages Ambiguity, Empathy, Decision Quality, Cultivates Innovation, Instills Trust, Strategic Mind and Insatiable Learning.

Read the list as a prep checklist rather than a slogan. Each item comes with a one-line definition on the page. "Manages Ambiguity" means operating effectively when the way forward is unclear, and "Courage" means stepping up to address difficult issues. Map a real story to each one. You will not need one story per trait, because a good story covers two or three of them, but you should know which story you will tell when an interviewer probes for any of them.

Can you use AI in a Snap interview?

Snap's How We Interview page states the default plainly. Snap wants to see your own thought process, so unless you are specifically told you are allowed, external resources, including AI, are not allowed during interviews.

Snap's technical hiring page adds the exception. It says your AI fluency may be evaluated in the process, that your recruiter will tell you in advance, and that your interviewers will clarify whether AI is permitted during the round.

The software engineering interview prep page shows where that happens. Each of its four tracks, Backend, Mobile, Full Stack and ML Infrastructure, lists an AI Assisted Coding round. Snap describes it as an evaluation of how you use AI as part of your engineering workflow. You may be asked to clarify requirements, make a plan, use an AI assistant to generate or revise code, debug, and validate the final solution. The page says strong candidates demonstrate intentional AI usage, sound engineering judgment, and the ability to explain, test and improve AI-generated output.

Read that description closely. It rewards the parts of the work that happen around the AI's output, such as the plan, the testing and the explanation, more than the output itself. Candidates who accept the first suggestion and move on are showing the opposite of intentional usage. Snap is not alone in splitting the rules by round. Applied Intuition runs a no-AI phone screen and then an onsite build where any AI model is allowed.

What technical rounds does a Snap software engineer face?

Snap's software engineering prep page breaks the rounds out by track, and the mix differs by track.

Backend. General coding, AI assisted coding, and system design.
Mobile. General coding in the language of the platform (Swift or Objective-C for iOS, Kotlin or Java for Android), AI assisted coding, mobile feature and application design, API and SDK design for more senior roles, and an iOS domain-specific round for iOS roles.
Full stack. Frontend coding and fundamentals, general coding, AI assisted coding, an interactivity and display round on building polished interfaces, and system design spanning front end to data layer.
ML infrastructure. General coding, system design, an ML infrastructure system design round, and AI assisted coding.

Snap's page says the general coding rounds assess data structures, algorithmic performance, object-oriented development and iterating on your code. System design can range from object-oriented design at junior levels to large-scale resilient systems. If you are preparing for the backend or full stack loop, work through a system design question set out loud, because the round is a conversation as much as a diagram.

What does the Snap product manager loop look like?

Snap's product interview prep page says the PM process gathers signals across four core competencies and Snap's three Values in Action, Kind, Smart and Creative. The four competencies are customer and product intuition, judgement and execution, strategic thinking, and cross-functional leadership and influence.

The same page says to expect a mix of behavioral and case study questions, usually an initial screen with the hiring manager followed by a virtual loop with four 1:1 interviews. Four interviews against four competencies is a tidy match, but Snap's page does not say each interview covers one competency, so prepare stories and cases for all four rather than guessing the split.

What is overrated in Snap prep

Memorizing STAR templates. Snap's How We Interview page tells you in writing which format it wants, and the Learning step is the one a STAR template leaves out. Candidates who recite polished STAR answers will sound prepared and still stop one beat short. The other overrated move is treating the AI rule as a trap to sidestep. Snap's technical hiring page says your recruiter will let you know in advance that AI fluency may be evaluated, and that interviewers will clarify whether AI is permitted in the round, so prepare for both modes deliberately.

How to prepare for the Snap loop

Rewrite five behavioral stories in S.A.I.L. form, with a Learning step that names a behavior you changed and a later moment you used it.
Map those stories against the values-in-action traits listed on Snap's How We Interview page, so you know which story answers which probe.
Ask your recruiter which rounds your loop includes and whether AI fluency will be evaluated. If you are unsure whether AI is permitted in a round, ask the interviewer at the start, which Snap's technical hiring page says interviewers will clarify.
Practice one AI assisted coding problem end to end. Write the plan before you prompt, test what the assistant returns, and explain every change you make to it.
Say your stories out loud before the interview. A voice mock interview shows whether the Learning step survives being spoken under time pressure, which is where it usually gets dropped.
Turn your camera on and check your setup. Snap's How We Interview page says virtual interviews run camera on.

Where this is heading

Snap's published process holds two positions at once. It bans AI by default because it wants to see your own thinking, and it builds a round to watch you think with AI when the rules allow it. The S.A.I.L. format asks for the same thing in behavioral form, which is evidence that you reflect on what you did. Candidates who can explain their own reasoning, with or without an assistant in the room, are the ones this process is built to find.

Originally published on the Four-Leaf blog.

The Applied Intuition interview process, with and without AI

Frank @ Four-Leaf — Thu, 01 Oct 2026 16:44:41 +0000

Key takeaways

Applied Intuition runs two interviews that pull in opposite directions. Its technical phone screen is a 45-minute coding task with no AI. Its onsite gives you a 45-minute system design question and then two hours to build what you designed with any AI model you want, followed by a demo to a panel of engineers. The company laid out that split in a July 2026 post on its engineering blog, written by software engineer Kesav Viswanadha, and the reasoning the company gives for each half tells you what each stage is for.

What are the stages of the Applied Intuition engineering interview?

Applied Intuition's July 2026 engineering blog post describes two technical stages for engineers, and it is explicit about which one changed.

The technical phone screen. "Keep our technical phone screen as is," the post says. "A 45-minute coding task with no AI." The company's stated reason is accountability. Engineers are still responsible for every line that goes into the codebase, so the screen exists to confirm a candidate understands code independent of any assistant.

The onsite. This is the stage the company rebuilt. It opens with a 45-minute system design question, an open-ended problem that the post says is usually derived from something the company has actually had to solve. Then the candidate gets two hours to build a working prototype of the system they just designed, using any AI model and any frameworks. Then comes the demo. Candidates present to a panel of engineers and walk through what they built, the decisions they made, and what they would do differently.

Applied Intuition's July 2026 engineering blog post covers engineering only. It does not describe recruiter calls, behavioral rounds, or the process for non-engineering roles, so ask your recruiter what else sits around these two stages for your specific loop.

Can you use AI in an Applied Intuition interview?

Applied Intuition's rules differ by stage, so the first thing to get straight is which rule applies where.

The phone screen is a no-AI coding task. Treat it like any traditional screen. Write the code yourself, talk through it, and expect your interviewer to judge the code itself.

The onsite build explicitly allows any AI model. Applied Intuition's July 2026 engineering blog post says the company retooled its whole organization around AI coding agents, that everyone there uses AI regularly, and that its old interview process, entirely no-AI coding and whiteboarding, was sending candidates a message that no longer matched how the company works. The two-hour build is how the company changed that.

The same Applied Intuition post also says the company is still scaling the new format across a company of over 1,000 engineers, and it says the format keeps changing because every model release shifts what candidates can do in two hours. Confirm the AI rules for each stage with your recruiter before the day. That is a routine question and a recruiter will expect it.

Companies publish different rules here. Waymo's published rules forbid AI in every live interview. Applied Intuition bans AI in its phone screen and allows any AI model in its onsite build.

What does Applied Intuition say its onsite surfaces?

Applied Intuition's July 2026 engineering blog post says AI fluency is important but only one aspect of the job. As the craft of coding gets easier, the post argues, decision-making, technical excellence and cultural factors matter more, and any format has to test those core skills with AI "just the medium to enable it."

In the company's trial of the new format, interviewers rated their confidence in seven hiring signals on a scale of 1 to 5, and six of the seven scored between 4.0 and 4.7. The post names practical skills, tackling unfamiliar challenges, technical communication, logical thinking and technical mastery among the ones the format surfaced clearly.

Applied Intuition does not say which part of the onsite measures which signal, so the rest of this paragraph is Four-Leaf's prep advice, and none of it is the company's rubric. Technical communication is on that list, and the demo is one moment the format asks you to explain yourself, walking a panel of engineers through what you built, the decisions you made, and what you would do differently. Prepare the explanation as seriously as the prototype.

Applied Intuition's post also reports a side effect it found after launch. Engineers train quickly to run the interview because it "mirrors a day at their job." Four-Leaf's advice follows from that. If you have never built anything with an AI coding agent, practice that before the onsite.

Why does Applied Intuition keep a no-AI coding screen?

Applied Intuition's July 2026 engineering blog post gives accountability as the reason. Engineers are still responsible for every line that goes into the codebase, so the company wants to know a candidate understands code independent of AI. The screen also has a second job. Of the seven hiring signals interviewers rated in Applied Intuition's trial of the new onsite, code elegance and design was the one that scored significantly lower, and Applied Intuition says that is "what we intend to continue evaluating with the phone screen."

That makes the screen the stage where Applied Intuition says it intends to keep evaluating code elegance and design. In the onsite build you may use any AI model you want. In the screen, nothing stands between your code and the interviewer.

Prepare for it like a traditional coding screen. Clean variable names, sensible decomposition, edge cases handled out loud, and a working solution inside 45 minutes. Candidates who have leaned on autocomplete for a year should practice a few problems with it switched off.

How did Applied Intuition's interviewers rate the new format?

Applied Intuition ran over 30 real onsite interviews with candidates as a trial and pulled in what Applied Intuition's July 2026 engineering blog post calls "the most vocally skeptical senior engineers" from across the company, including people who wrote the previous generation of interview questions.

The results come from 15 unique interviewers across dozens of responses. 86% thought the format gave them a better read on candidates' ability to work there. 85% said it was better than the old format. And 91% reported they were more confident in their scorecard decision.

Read those figures carefully. These are interviewer perceptions from a trial of about 30 candidates, published by the company that designed the format, so they measure how confident the panel feels. They say nothing about whether the hires turned out better.

Does Applied Intuition value founder experience?

It says so plainly. In a 2024 company blog post on former founders at Applied Intuition, co-founder and CTO Peter Ludwig said the company had sought them out in recruiting. "Seeing on the resume that somebody started something, we consider that a very positive signal." The company's careers page describes a team that includes 40 ex-CTOs and 30 former founders.

If you have ever started something, put it on the resume and be ready to talk about it, even if it failed. Taking an open problem from a design conversation to a working demo in two hours is also, in Four-Leaf's reading, close to what founders do.

What is overrated in Applied Intuition prep

Grinding a hundred LeetCode problems for the onsite. Applied Intuition replaced an onsite that was entirely no-AI coding and whiteboarding, and a recent hire quoted in Applied Intuition's July 2026 engineering blog post called the new format "so refreshing and much better than recycled leetcode questions." Algorithm drilling still earns its keep for the phone screen. Past that, your hours are better spent building small systems end to end with an AI agent and explaining them to someone.

How to prepare for the Applied Intuition loop

Ask your recruiter which stages your loop includes and what the AI rules are for each. The format is new and still being scaled.
Practice timed coding problems with AI tools switched off. The phone screen is 45 minutes, unassisted, and Applied Intuition says it is where it intends to keep evaluating code elegance and design.
Build two or three small prototypes from a blank repo in two hours each, using the AI coding tools you would actually reach for. Time yourself from design to working demo.
Rehearse the demo out loud. Walk through what you built, why you made each decision, and what you would change. Applied Intuition's post says candidates walk the panel through exactly those three things, and a voice mock interview is a cheap way to hear how your explanation lands before a panel of engineers does.
Read the system design rubric breakdown and practice talking through an open-ended design in 45 minutes, since that design is what you will then have to build.

Where this is heading

Applied Intuition's loop, as its July 2026 engineering blog post describes it, is a clean statement of one position on AI in hiring. Check the fundamentals without AI, then watch how the candidate works with it. Four-Leaf's read is that the candidates best placed for a loop like this can write code alone, can build with AI tools when the rules allow them, and can explain every choice to the engineers deciding whether to trust them.

Originally published on the Four-Leaf blog.

Is AI taking entry-level jobs? What the hiring data shows

Frank @ Four-Leaf — Wed, 30 Sep 2026 16:44:52 +0000

Three studies published in 2026 now measure what AI is doing to the bottom of the career ladder, and they agree on something narrower than the headlines. Companies that adopt AI are adding senior people faster than junior ones, and young workers in the most exposed occupations are being hired less. None of the three found a wave of entry-level layoffs.

The biggest of the three is a Stanford Digital Economy Lab working paper by Bharat Chandar and Bouke Klein Teeselink, published September 21, 2026. It analyzes 1.25 billion job postings and 154 million employment records across 41 countries, and its abstract says the fall in the junior share of the workforce "primarily comes from growth in senior employment rather than from a fall in junior employment."

That distinction decides what an early-career candidate should do next, so it is worth reading the evidence slowly.

What did the 41-country Stanford study find?

It found that AI adopters grow top-heavy. Chandar and Klein Teeselink compare foreign affiliates of multinationals that adopted generative AI with matched affiliates that did not, in the same country and industry and of similar size before ChatGPT launched. In Chandar's own summary, adopting affiliates saw senior employment rise 6.7% relative to those matches, which "adds up to a two percentage point decline in the share of employees who are junior."

Overall employment at adopters rose about 3.3% relative to matched non-adopters, a result Chandar calls marginally statistically significant. The junior share fell in 23 out of 31 countries where the team could get precise estimates, and the decline was statistically significant in seven of them.

Two limits travel with the finding. The comparison is adopters against non-adopters, so it describes how AI reshapes a firm, and Chandar writes that this tells us little about total employment across the economy. Adoption is measured from job postings that mention generative AI, which the authors treat as "a lower bound on AI adoption."

The paper's abstract attributes the shrinking junior share primarily to growth in senior employment. Seniors are being added, and the junior slice shrinks as a share of the whole.

Are young workers in AI-exposed jobs being hired less?

In the US payroll data, yes, and the gap is widening. The Stanford Digital Economy Lab's August 2026 update of "Canaries in the Coal Mine?", by Erik Brynjolfsson, Chandar and Ruyu Chen, uses ADP payroll records. It finds employment among workers ages 22 to 25 in highly AI-exposed occupations now about 19% below where it would be had it kept pace with same-age workers in less-exposed occupations. At the July 2025 data vintage that shortfall was 15%.

Measured in levels, the same update reports that employment of workers ages 22 to 25 in the two most exposed quintiles fell about 11% between November 2022 and June 2026, while the same age group in the three least-exposed quintiles grew about 10%. "Experienced workers show no comparable gap," the authors write.

The mechanism matters most for a job seeker. The canaries update says the adjustment "appears to operate primarily through reduced hiring of young workers rather than increased separations." People already in these jobs are mostly keeping them. Fewer new ones are being let in.

The authors are careful about cause. They call these "descriptive patterns, not causal estimates" and note that the gaps between more- and less-exposed young workers "shrink when we account for education." Several alternative explanations, including interest rates and remote work, did not appear sufficient to account for the pattern in their tests.

Are there fewer entry-level job postings?

In the most exposed fields, entry-level roles are a much smaller share of what gets posted. Indeed Hiring Lab's September 2026 analysis of postings that advertise an annual salary found the entry-level share in the most AI-exposed occupations fell from 29% to 10% between 2021 and 2026.

Read that figure on its own basis. It covers salaried postings only, a population Indeed says skews senior (about 37% senior by 2026, against about 14% across all US postings), and it is a share of postings, never a count of jobs. The pay side of the same study, and why it points the other way from the fear, is covered in our breakdown of AI exposure and pay.

Four-Leaf's own index shows the same shape without any AI lens at all. Across the active postings in the June 2026 snapshot, classified by title, senior-and-above titles made up 26% against 9.0% for intern, junior, entry-level, new-grad and associate titles. That is a single snapshot, so it cannot say whether the tilt is new or what caused it. The full write-up of that index has the rest of the numbers.

Indeed's and Four-Leaf's figures each show the entry tier thinner than the senior tier, on different bases, so do not add them together.

Is AI taking entry-level jobs or changing who gets hired?

The evidence so far says AI is changing who gets hired more than it is removing people already hired. Chandar put his own reading plainly: "no widespread displacement so far, likely decently-sized negative impacts for exposed junior workers."

For a candidate, that difference is practical. A layoff wave would mean the jobs are gone. A hiring tilt means entry-level roles still exist but make up a smaller share of hiring in exposed fields. That is a targeting problem, and targeting is something a candidate controls.

The canaries update also splits occupations by the kind of knowledge they run on. It found employment falling among young workers in occupations that rely heavily on codified knowledge, the formal, documented kind taught in textbooks, and rising among experienced workers in occupations that lean on "tacit knowledge acquired through practice, mentorship, and repeated exposure to real situations." That describes occupations, not individual candidates, but it is a useful frame for what to put in front of a reviewer.

What is overrated

The advice to avoid AI-exposed fields entirely is overrated. The Stanford canaries update found the declines concentrated in occupations where AI tends to automate the work, while "in occupations where AI is used more to complement workers, employment is flat or rising, particularly among more experienced workers." Exposure alone does not tell you which of those a field is. Walking away from a field because a chart moved is a bigger bet than most early-career candidates realize.

Volume is overrated too. When the entry tier is a small share of postings, sending two hundred identical applications mostly means competing in the same crowded slots two hundred times.

What to do this week

Search by level on purpose. In Four-Leaf's AI job search, a level word like "junior" or "entry" in the query gives titles that carry junior, entry or associate a ranking boost, so they rise, without dropping postings whose title carries no level word. When the search names a company, that company's roles are pinned first and sorted by role match instead. Many real starter roles never put a level in the title.
Lead with work you have done. A project, an internship deliverable or a shipped fix is evidence of practice. Our read is that evidence of practice travels further than a list of courses.
Write for the level above. Describe your strongest piece of work in terms of the decision it informed or the problem it closed, the way a hiring manager would describe a mid-level hire's work.
Cut the list, deepen each application. Pick fewer roles, read each posting closely, and rewrite your bullets in that posting's language rather than sending one resume everywhere. Four-Leaf's resume builder tailors a resume to a specific posting.
Keep applying to exposed fields where you have proof. Entry-level roles are a smaller share of postings there, and still present.

Where this is heading

The canaries authors say plainly that no single study settles this, and their dashboard will update monthly. The trend could accelerate, stall or reverse. What 2026 has already shown is where the pressure lands. It lands on the first hire. In exposed fields the entry-level job is still there, in smaller numbers, and our bet is that it goes to the applicant who shows what they have already done.

Originally published on the Four-Leaf blog.