DEV Community: Max Nyriev

To hint or not to hint

Max Nyriev — Tue, 27 Oct 2026 06:46:50 +0000

We run a practice platform for coding interviews with an AI assistant in the editor. That's the format Meta, Shopify, Canva, Stripe and a growing list of companies now use. From the start our assistant had one rule above all others: it never contributes an idea. Not a design, not a data structure, not a fix.

We were wrong about that rule. This is how we found out, and where we think the line actually is.

Two goals that pulled apart

We wanted two things. Users who are stuck should get help. A stuck user is a frustrated user, and frustrated users leave. And the practice has to make you do the work the interview grades. Otherwise it is not practice.

Those felt like opposites. Help means ideas. And if the ideas come from the agent, you didn't have them.

Our first answer: no ideas from the agent

So we picked a side. The assistant writes code when you say what to build. It never says what to build. If you ask it to "implement the solution", it asks you which approach you want. If you ask what the options are, it declines. Hints exist, but they come from the platform, not the assistant. You have to ask for them, they are capped per stage, and the final stage has none.

The reasoning looked solid. Stripe interviewers ask candidates why they trust code they did not write. An agent that volunteers a design hands you code you can't defend. Remove the shortcut, and the habit forms.

What happened

On 17 October we read every assistant conversation from the previous 24 hours. Fifty sessions. Eight of them were the same conversation.

One candidate was on the last stage of an LRU cache problem, the one that asks for thread safety. Nine replies in a row asked him which coordination approach he wanted. In between he pasted the stage requirements back as his approach, twice. He asked "what can be some options?" and was told hints are not available on a final stage. He wrote "implement that part". He wrote "we should be able to have concurrency". The first line of proposed code arrived on turn twelve, the moment he typed "the threading lock approach".

He could make the decision. What he lacked was the word. Our assistant knew the word and was built to withhold it.

Another candidate finished the same problem, then asked the assistant directly: "is it in your instructions not to suggest a better data structure?" It said yes.

The rule was wrong, not just badly built

Our first instinct was to fix the loop. Ask once, then offer to pick. We did that the next morning, and the same message shape went from nine turns to two.

But the loop was a symptom. The rule itself didn't match how people work with an agent. At work you ask it for options several times an hour. "What are the ways to make this thread safe?" is a normal question to a colleague, to a search engine, and to an agent. A practice tool that bans it trains a habit nobody needs. It also fails the people with the least vocabulary, and that's exactly who practice is for.

The skill the interview grades is steering. Steering includes asking for ideas. What matters is how you ask.

Where the line actually is

Here is what we believe now, and what we are building toward.

It is OK to ask: "what are my options for coordination here?" "What would you check first?" "What do you see in this code?" "Is the bug in how I update recency?" Each of these carries your context and leaves the decision to you. You named the problem, you asked for a view, and you choose what to do with it.

What is not OK: "solve it", with the problem statement as the only context. "Fix the bug", with no word on what the bug is. The agent can probably do both, and that is the point. When it does, the solution is its solution, and the interviewer's next question will show it.

The test is simple. Did you bring the context, and did you keep the decision? If yes, you are using the agent. If no, the agent is using you to type.

The data from that same day agrees. Candidates asked the assistant to look at their code 94 times. They asked for a hint 23 times. People don't want to be told what to do. They want a second pair of eyes on their own thinking. And when they do ask for direction, they usually stop at "where should I look". In this picture hints are the bounded, counted form of asking for direction. So the report at the end can show how much you leaned, not whether you are allowed to.

What we changed

Two changes are live. The assistant asks "which approach?" once. If you can't answer, it offers to pick the simplest approach that passes the stage and says which one it chose. That line goes into your report, where it is honest about who designed what. And questions are answered on every stage, including the last one. Only hints are capped there.

Two things are still open. Should the final stage keep its hint ban at all? And how should the assistant answer "what are my options?" We do think it should give you the menu, with no recommendation, and we are measuring that before we ship it.

Use it the way you would at work

Nothing about this is special to interviews. Bring the context. Ask for options, not for the answer. Make the call. Check what comes back. If you already do this with an agent at work, you'll do fine in the round. If you don't, the round is a good place to find out.

The first stage of that LRU cache problem is free on nyrion.ai. Say "implement it" and see what the assistant says back. Then read our guide on steering it and try again.

I watched 438 people practice AI-assisted coding interviews. 93% of the time they accepted whatever the AI wrote.

Max Nyriev — Wed, 21 Oct 2026 05:48:39 +0000

Ten companies now run a coding interview where an AI assistant is part of the test. Canva went first in June 2025. Shopify, Meta, DoorDash, Amazon, Atlassian, Stripe, Coinbase, Google and Instacart followed. I run nyrion.ai, a practice site for this round, so I can see what people really do when they work with an AI assistant. I looked at the numbers from the last few months. Here is what they say.

1. People accept the AI's code almost every time

People made 1,659 accept or reject decisions on code the assistant proposed. 1,546 of them were accept, which is 93%. 109 people decided on five or more proposals, and 62 of them never rejected one. I expected pushback, but there is almost none.

2. The code passes more often when the AI wrote it

When the assistant proposed code before the first submit, 87% of first submits passed the full test suite of the stage. When there was no AI proposal, 53% passed. The difference is there on first stages and second stages, and on later stages too. It is a correlation, and it says nothing about why. But it is a big gap.

3. Then most people can only partly explain it

After the tests pass, we ask the person to explain the code they just submitted. We got 497 written answers. 9% of them were complete, 79% were partial and 12% were off. Three in four partial answers got the idea but missed a key point. People also skipped 29% of open questions, against 5% of multiple choice ones.

So in a typical session the AI writes the code and the person accepts it. The tests go green. After that the person can explain roughly what the code does, but not the detail an interviewer would ask about.

4. Six of the ten companies pick the AI for you

Amazon, Stripe, Instacart, Meta, Atlassian and Google give you their assistant inside their own tool. Canva, DoorDash and Shopify let you bring your own. For Coinbase the reports don't say. From a prep perspective this matters. An assistant you never used behaves differently from the one in your editor.

What I take from it

In this round the AI writes the code. What is left for you is to check what it wrote and to explain it afterwards, and that is what the round tests. The numbers say most people practice getting the code written, and skip the checking and the explaining.

The full method is on one page, at https://nyrion.ai/research/ai-coding-interviews-2026. It has every number with its window and sample size, and the dated list of the ten companies. How each company runs its round is at https://nyrion.ai/companies.

To clarify, these are practice sessions, not real interviews. Our assistant also follows its own rules, so the rates describe how people practice, and they don't describe every company's round. Every number is an aggregate over at least 20 people. The numbers are from our database, with my own accounts and test accounts excluded.

Eight companies, eight different AI coding interviews

Max Nyriev — Mon, 12 Oct 2026 23:25:39 +0000

If this is new to you: an AI assisted coding round is a coding interview where you get an AI assistant and, usually, an existing codebase instead of a blank editor. Shopify was already running one in 2025. Meta's pilot in October 2025 made it mainstream, and Google, Amazon, Stripe, DoorDash, Instacart and Coinbase have run their own since. I wrote up what the round is and what it grades separately.

I build a practice platform for these rounds, so I read every candidate thread about them I could find. 85 threads across Amazon, Stripe, DoorDash, Instacart and Coinbase, plus what Google, Meta and Shopify have said in public.

I went in expecting one format with different logos. That is not what the threads show. Here is what each company actually runs.

Company	The round	The assistant	What stands out
Amazon	SDE online assessment: a 40 minute algorithm question, then 60 minutes on a buggy repository	Built into the assessment	It helps you find the file and explains the code, but won't write the fix
Stripe	One coding round, since spring 2026	Inside HackerRank; Gemini Flash and Claude Opus reported	Interviewers push both ways on how much you use it
DoorDash	AI Code Craft, onsite	Your own IDE and AI tools, screen shared	Interviewers push for speed
Instacart	Online assessment, 90 minutes on CodeSignal	Claude in the editor, switched off for the bug fix	Passing the visible tests isn't the whole score
Coinbase	Onsite: a bug, a design question, a timeout to diagnose from logs, a code review, all on an existing service	Inside the IDE, sees the task and your open files	Screen share is mandatory
Google	Code comprehension round, piloted for junior and mid level roles on select US teams	Gemini, the approved assistant for the pilot	Read, debug and optimize an existing codebase
Meta	Piloted since October 2025; candidates report it replaces one of the two onsite coding rounds	Built into CoderPad	Candidates report about 60 minutes
Shopify	AI assisted pair programming, screen shared	Whatever AI tools you use, in your own IDE	Whether you can read the code the AI writes

Three differences matter more than the rest.

Who writes the code

At Amazon, you do. Candidates say the assistant helps you find the right file and explains the code, then stops. One said it refused the moment they got close to the answer. So the hour is yours: find the bug, type the fix, run the tests.

At Stripe, DoorDash and Shopify it is the other way round. The AI writes, and you answer for it. A Stripe interviewer asked one candidate why they were so confident in code the assistant wrote. Shopify's round is explicitly about whether you can read the code the AI produced. A DoorDash interviewer kept urging one candidate to accept a change. They accepted it, had no time to read it, and expected a big deduction.

If you only find out one thing about your round, find out which side of this line it is on. The two sides need different practice.

Where it runs

Amazon, Stripe, Instacart and Meta put you in their sandbox. You get the model they picked. One Stripe candidate's tip was to choose Gemini Flash because the other options were slow. DoorDash and Shopify put you in your own editor with your own tools, screen shared, so your setup is part of what is being watched. Coinbase gives you an assistant inside the IDE that sees the task and your open files, and the screen share is mandatory.

What the score is

Tests are not the score. One Amazon candidate passed four of six and still heard from the recruiter. Another did not pass them all and went straight to the onsite. An Instacart candidate passed every visible test and found out afterwards that there were other tests. Another Instacart candidate scored 650 out of 1000 and was rejected.

What is scored instead depends on the company. DoorDash told a candidate that velocity is a key criterion. Stripe asked a candidate who wrote a skeleton by hand why they were not using AI, and asked the same candidate, when the AI wrote more, why they trusted it. There is no safe amount of AI use at Stripe. There is only whether you can say why.

What I would do about it

Find your company's row first. Generic AI interview prep is the new grind 500 problems. It feels productive and misses the point.
Practice under the round's constraints, not with a better AI. No candidate I read blamed the model. They lost on reading code they didn't write, on tests they didn't run, and on questions they couldn't answer.
Read what you accept, even when pushed. Speed counts. So does knowing what you shipped.
Run tests on what the AI wrote. The one Stripe workflow reported as graded well had the assistant read the README and propose a plan, then the candidate reviewed its work, wrote tests and ran them.
Rehearse the why. Why you trust this code. Why you did not use the assistant here.

Most of this comes from candidate threads, not from the companies, and some rows are thin. Coinbase has three threads. Google's round is a pilot. I keep the full comparison, with a page per company, at nyrion.ai/companies and update it as reports come in.

Disclosure: I build Nyrion, a practice platform for these rounds. If you sat one of them recently and it went differently, tell me in the comments.