DEV Community

Harpreet Sahota
Harpreet Sahota

Posted on Originally published at employabledatascientist.com

AI Didn't Kill the DevRel Interview. It Made the Old Tricks Load-Bearing.

#devrel #ai #career #freelance

AI Didn't Kill the DevRel Interview. It Made the Old Tricks Load-Bearing.

TL;DR

  • DevRel job posts in late October 2026 don't screen for a resume, they screen for a public body of work: Chalk wants "a body of technical content you can point to... send the links," LightSprint wants "evidence: a community you grew, a following."
  • The live interview round didn't get easier with AI in the room. DoorDash, Google, and Cursor now score how you direct and narrate an AI agent, which is the same "process" signal Evan Pellett coached candidates to perform back in 2020, just graded live instead of described after the fact.
  • LiteLLM, bem, Lightsage, Nango, and LightSprint are all hiring their first dedicated DevRel person right now, which means your first year is Jacqueline Nolis and Emily Robinson's "first data scientist" problem: define the job, do the job, and set the precedent, without getting stuck building things nobody asked for.
  • When you're ready to go independent, the data backs you: AI-tagged freelance work pays 2.5x non-AI work and 77 percent of business leaders say the AI era is increasing their need for fractional talent, which is Alison Grade's freelance math with better numbers behind it.

Abstract

Three old episodes of this show keep answering questions I'm getting from devrel engineers in 2026: an interview-science recruiter, two data scientists who wrote the book on building a career in the field, and a freelancer who wrote the actual Freelance Bible. None of them said a word about AI developer tools. All three describe exactly what's happening in devrel hiring, leveling up, and going independent right now. This post pulls the specific, checkable nugget from each episode, ties it to a 2026 news item with a date and a link, and gives you one move per career stage: getting the devrel job, surviving the first year of it, and leaving it on your own terms.

Table of Contents

  • Standing Out: The Interview Became a Live Demo of Your Judgment
  • Leveling Up: You're the First Data Scientist Again
  • Breaking Free: Pricing the Thing Nobody Else Can Verify
  • Summary
  • Next Steps
  • Resources

Standing Out: The Interview Became a Live Demo of Your Judgment

In 2020, Evan Pellett told me the real exam in a job interview isn't the questions, it's the unconscious ones the hiring manager never says out loud: will this person perform, do they have the energy to last, can I actually see how they think. His framework, REAP, put Results first. A hiring manager spends the interview hunting for proof you've already done the job, and if you don't hand it to them directly, they'll spend the rest of the hour trying to extract it through trivia.

Six years later, "process" stopped being something a candidate describes after the fact.

DoorDash now runs its engineering interview as a 60-minute AI-assisted working session: your own IDE, your own coding agent, a task shaped like real DoorDash work. DoorDash scores five things: how you get oriented, how you use AI and verify its output, how you debug, how you manage scope, and how you communicate tradeoffs. Finishing every task matters less than showing a tight loop and good judgment. In the second half of 2026, Google started permitting an approved AI assistant in its "code comprehension" round, where candidates read, debug, and optimize an existing codebase. Cursor, per the live AI Interview Index, doesn't run a single non-AI coding round anymore. The whole loop assumes you build with an agent, because that's the job now.

That's Pellett's Process pillar, graded in real time instead of recalled in a story. The failure mode he named in 2020, "subtle defiance," the candidate who won't go deep, who gives a vague non-answer instead of walking you through the reasoning, now shows up the moment you go quiet instead of narrating what you're trying, why it failed, and what you're trying next. An interviewer watching you direct an AI agent can see hesitation and avoidance just as clearly as a recruiter could see it across a table in 2007.

The harder part happens before you ever get a live round. Read the actual job posts going up for devrel roles at AI companies right now. Chalk, hiring a senior devrel lead, asks for "a body of technical content you can point to, especially video and live demos. Send the links," and for someone who already "build[s] the way your audience builds: Git-driven CI/CD, CLIs, notebook to production, coding agents like Claude Code, Codex, and Cursor in your daily loop." Lightsage wants someone "technical enough to build demos, read API docs and run coding agents yourself," with "strong technical writing and a public portfolio." LightSprint is blunter: "you're a builder of audiences. You have evidence: a community you grew, a following, a channel, a newsletter, something that shows people choose to listen to you."

That's Pellett's "string of results," the one he coached candidates to pull from "the first time in high school right up into the current." In 2026 the proof isn't a bullet on a resume, it's a public repo, a YouTube walkthrough, or a blog post someone else can click on before they ever call you. The application happens months before the req goes up. The live round just confirms it by watching you think out loud with an agent in the room.

If you're on the other side of this, hiring instead of applying, the founder-facing version of the same scorecard is the first DevRel hire should not be a megaphone.

Leveling Up: You're the First Data Scientist Again

Look at who's actually posting these devrel roles. LiteLLM calls it "our first dedicated Developer Relations role." bem says "we are hiring our first Developer Relations hire." Lightsage and Nango both say "Founding." LightSprint says "this is our first community and developer relations hire." Five different AI devtool companies, same sentence, same month.

Landing the job is the easy part. The hard part is that you inherit nothing: no playbook, no precedent, no second opinion from someone who's done this role here before you.

Jacqueline Nolis has been the first data scientist at three or four companies, depending how you count. She described the job as three things fighting for the same hours: figuring out what the job even is, actually doing the work, and setting the precedent everyone hired after you will inherit. Get stuck defining the role and nothing ships. Get stuck shipping and the team that joins later has no standards to build on. Get stuck setting precedent and the business never sees enough value to hire a second person. The job is holding all three at once, continuously, not picking one.

The specific trap she named for data scientists maps onto devrel without changing a word. "I think probably about 80 percent of the work that data scientists do is not useful to anyone... as a field we haven't really grappled yet with how do we quit early." She calls the failure mode "ivory tower," work that's intellectually interesting to build and that nobody asked for. For a founding devrel hire, ivory tower looks like a beautifully produced sample app nobody clones, a newsletter with declining opens, a conference talk prepared for an audience that isn't your actual developer. It feels like output. It isn't precedent.

The 2026 research on DevRel career ladders backs her up from the other direction. Every DevRel lead interviewed for that research said the same thing: promotion is recognition of work already done at the next level, never an incentive for potential. The mechanism it recommends, a running evidence log plus a "difference doc" for what separates your current level from the next one, exists because most devrel work happens outside a manager's direct view. That log is the same artifact Nolis describes as setting the precedent. If you're not writing it down as you go, you can't prove you did it, and neither can the person who joins after you.

Three things keep you out of the ivory tower quadrant as a founding hire: a plan with a day-90 outcome instead of a wish list (see the 30/60/90 day devrel plan), a weekly rhythm light enough to actually run (see the practical devrel operating system), and a metrics report that separates real signal from noise before you hand it to a CFO who's going to ask (see community metrics that survive CFO review). None of that is exciting work. It's precisely the kind of unglamorous process Nolis says most data scientists skip, and it's the difference between a founding devrel hire who gets backfilled into a team and one who gets quietly let go when the budget tightens.

Breaking Free: Pricing the Thing Nobody Else Can Verify

Alison Grade wrote The Freelance Bible around one idea: a freelancer stands on a three-legged stool, financial drive, skills, and desire, and if any leg goes missing you don't have a business, you have a hobby. Take away financial drive and you've got a side project. Take away skills and you've got enthusiasm with no substance behind it. Take away desire and you can't communicate the value of what you actually know.

Her other rule is simple, and most new freelancers break it first. Price on the value you deliver, not the hours you log, and if work is coming out of your ears, you're not charging enough. She also insisted on writing her own client brief rather than waiting for one, because "my client's not going to be able to write the brief in the proposal because they don't understand what I can deliver."

The 2026 numbers make that advice sharper than it was in 2020. AI-tagged freelance work now pays 2.5x more than non-AI-tagged projects, with AI consulting up 24 percent quarter over quarter. On Upwork, AI strategy consulting grew 51 percent year over year in Q2 2026 while the marketplace itself shrank 4 percent, and the same report found nearly half of freelancers said their most recent job was AI-related, while only 16 percent of the job postings actually said "AI" anywhere in the text. Clients don't know to label the work. The freelancer who wins the proposal is the one who names the tool, the workflow, and the measurable result, exactly Grade's advice to write the brief yourself instead of waiting for a client who doesn't know how to write one.

The demand side backs up what Grade predicted in the middle of a different crisis, back in 2020: "I think we're going to see huge numbers of freelancing opportunities... people will be more risk averse in terms of taking people on." Swap coronavirus for AI headcount anxiety and the mechanism is identical. 77 percent of business leaders now say the AI era is increasing their need for fractional talent over full-time hires, and Robert Half found 29 percent of hiring managers have already reopened roles they cut after an AI rollout, this time as contract positions. A devrel engineer who spent two or three years as a founding hire, building the exact evidence log the career-ladder research recommends, is sitting on precisely the proof an AI-era client can't fake and can't easily hire for on a full-time line.

Want help figuring out whether your devrel skillset is ready to go fractional, and how to price it? That's a retainer conversation, not a resume review.

Summary

Pellett's REAP framework, Nolis and Robinson's "first data scientist" triage, and Grade's three-legged stool were never really about data science careers or interview scripts. They're the same three problems every devrel engineer in AI developer tools is facing on a different timeline: prove you've already done the work before anyone interviews you, survive being the precedent-setter when nobody else has done this job here yet, and know your own value well enough to charge for it once you're free to set the price yourself.

Next Steps

  • If you're applying: audit your own public work for Pellett's "string of results" before you touch a resume. If a stranger can't click through to something you built, you don't have a portfolio yet.
  • If you just took a founding devrel role: write your 90-day plan and your evidence log in the same week. Don't wait for a manager to ask for either.
  • If you're weighing freelance: run your current rate against the 2.5x AI-freelance premium and Grade's three-legged stool before you quote anything.

Resources

Top comments (0)