June's HN "Who Is Hiring" Killed the Coding Screen. AES Wrote It Down.
The June 2026 HN "Who Is Hiring" thread explicitly demotes raw coding. Here's how to source the "direct the AI" talent LinkedIn can't see.
If you sourced for May's HN "Who Is Hiring" thread, you noticed Cursor and Claude Code showing up as a "nice to have." The June 2, 2026 thread crosses a line May only hinted at. Two posts say outright that raw coding ability is no longer the bar, and the actual screen is whether you can direct an AI at production code and catch it when it lies.
That sentence breaks the way most recruiters source. The job title doesn't exist on LinkedIn. The skill tags aren't standardized. The signal lives in cover notes, CLAUDE.md files, and GitHub commit history. If you're still pasting "AI Engineer Claude Cursor" into LinkedIn Recruiter, you are looking at four people in the United States. The pool is much larger than that. It's just mislabeled.
What AES actually wrote
The cleanest example is AES, Associated Environmental Systems, an environmental-test-chamber manufacturer out of Chelmsford and Waltham. They posted a Summer Software Engineering Intern role in the June thread. Remote US, $30 to $40 an hour, June 15 to August 22. The JD opens with this:
The senior lead doesn't write code from scratch. Neither will you.
Then, a few lines down:
The bar is NOT raw coding ability. We use Claude Code for all code generation. The bar is the ability to direct an AI at production engineering and catch it when it's wrong: schema literacy, diff-reading, end-to-end ownership.
The stack is specific: Next.js 14 App Router, RSC, Server Actions, TypeScript strict, Prisma and Postgres, Tailwind and shadcn/ui, NextAuth, Twilio, the Anthropic SDK. The must-haves include "a link to AI-built work, a PR, deployed project, or repo, ideally with a CLAUDE.md" and "at least one shipped end-to-end project a stranger can use." And then: "No leetcode. No whiteboard."
This is not a startup writing cute copy. AES makes industrial hardware. They're building an internal field-service-management platform, and they've decided the cheapest way to staff it is to find an intern who can review Claude Code diffs faster than a senior would write the diff themselves.
Catalyst, Railway, Axis: same week, same pattern
AES is not alone. Catalyst·Wayfare AI (apply: talent+hn@cwai.co) posted in the same thread with Claude, OpenAI, Gemini, and Perplexity APIs in production, Cursor and Claude Code as daily tools, evals as first-class artifacts. Their screening line is doing a lot of work:
Short cover note matters more than a resume.
Railway (3 million plus users) opens its JD with rhetorical questions about wanting Claude to manage your infra, then frames the work as "distributed systems engineering, interface design, and agentic computing." Axis Communications, the video-management R&D shop, posted a role for a newly formed team where Claude Code, Copilot, Codex, and emerging agent workflows are "central to how the team operates" rather than a side experiment.
Four serious posts, one week, all saying the same thing in different registers: we hire on AI direction, not on raw code production.
May was about tool mention volume. June is about tool gating.
Why your LinkedIn search returns nothing
Here is the part that should worry you. We ran the obvious queries against our index.
Broaden it to a free-text headline search for "Claude Code Cursor AI engineer" across any title in the US and you get 16 people. Sixteen. For a category that, by external evidence, has tens of thousands of qualified practitioners.
Why the gap? Three reasons:
- Title drift. The role described in these JDs is a senior IC who reviews diffs and owns end-to-end shipping. Nobody has "AI Director" on a real builder's LinkedIn. The Boolean returns consultants and middle managers. The actual practitioners are titled "founding engineer," "software engineer," "intern," or nothing at all because they're independent.
- Skill-tag lag. LinkedIn's skills graph approves new tags in waves. "Claude Code" was not a standardized skill until late 2025, and most engineers don't go back and retag.
- The signal isn't in profiles. It's in artifacts. AES says it explicitly: send a CLAUDE.md, a deployed app, a PR. None of that lives on LinkedIn.
This is the trap. The talent pool isn't small. It's invisible to the tool you're using.
What the new screen actually is
Read the AES and Catalyst JDs side by side and the pattern is clear. The new screen is not a skill. It is two artifacts:
- A deployed project a stranger can use.
- A CLAUDE.md (or equivalent agent-instruction file) checked into a public repo.
A CLAUDE.md is a Markdown file that lives at the root of a repository and tells Claude Code how to behave inside that codebase. Conventions, file layout, what to test, what not to touch. It is, by accident, the cleanest possible proxy for the "direct the AI" capability AES is screening for. You cannot fake one. Either you've shipped enough Claude Code work to know what belongs in it, or you haven't.
Forrest Chang's CLAUDE.md repo, born from Karpathy's January 2026 viral post on AI coding pitfalls, hit 144,000 GitHub stars in weeks. A single Markdown file with zero runtime dependencies. That's the audience size. The 4dayweek.io board now lists 129 open roles requiring Claude Code from entry-level through staff. The labor market is real. The directory is GitHub, not LinkedIn.
Sourcing the invisible pool
So how do you actually find these people if Boolean is dead?
Search the artifact, not the profile
The single highest-yield query a sourcer can run in June 2026 is a GitHub code search for filename:CLAUDE.md filtered by recent commits and language. You get the writers. You get a sense of their style. You get repos you can read end-to-end before the first outreach. Stars on the repo are a noisy signal; recency and specificity of the CLAUDE.md are not.
This is exactly the friction that pushed us to build Refolk. You describe the person ("US-based engineer who has shipped a Next.js app on Vercel and maintains a non-trivial CLAUDE.md in a public repo, no big-tech titles required") and you get a ranked shortlist across GitHub, LinkedIn, and the open web in one pass. The CLAUDE.md signal is the kind of thing a keyword field can't represent but a plain-English brief can.
Treat cover notes as the work sample
Catalyst·Wayfare's line, "Short cover note matters more than a resume," is not politeness. It is the screen. The written-comms test, how you describe what you built and why, is the proxy for the diff-reading, "catch the AI when it's wrong" capability. Public writing is testable: Substack, README files, PR descriptions, conference CFP abstracts. None of it is indexed by LinkedIn. All of it is indexed by Google and by us.
Ignore the "AI Director" title entirely
If a hiring manager hands you a req for "AI Director" or "Head of AI" and the actual JD reads like AES's, push back. The title is poisoning the funnel. Search for the artifacts and the outputs, then map them onto whatever title the comp band requires at offer stage. The people doing this work right now mostly call themselves "founding engineer" or "software engineer" or nothing.
Read GitHub commit history like a resume
Frequency, scope, and the ratio of human-written to agent-generated commits is, in 2026, a more honest signal than years of experience. Boris Cherny, the creator of Claude Code at Anthropic, said in a recent interview that he hasn't written a line of code himself in more than six months, that coding is effectively "solved" for his kind of work, and that the title "software engineer" could start disappearing by the end of this year. He also mentioned that 20 percent of Anthropic's engineers began using Claude Code on the first day it was released internally. If the company that built the tool is at that adoption rate, your candidates' commit graphs should look different than they did in 2024, and that difference is the signal.
What this means for your June pipeline
If you run engineering sourcing, three concrete changes for the next 30 days:
One. Stop screening on Claude or Cursor skill tags. They are saturated where they exist and missing where they matter. Screen on artifacts: a CLAUDE.md, a live URL, a public PR description that shows diff-reading judgment.
Two. Rewrite your outbound first message. "I saw your Claude Code experience on LinkedIn" is dead on arrival because LinkedIn is not where the signal lives. "I read your CLAUDE.md in [repo] and the way you handle [specific convention] is exactly what we need for [specific problem]" gets replies. This is the kind of pre-research that gets faster when sourcing AI engineers is structured around the artifact, not the title, which is one of the workflows Refolk is built to surface in a single query.
Three. Brief your hiring managers on the AES line directly. If they want the role AES described, the intake conversation needs to switch from "five years of Python" to "show me the last three PRs you'd accept from this candidate." It is a different conversation, and it is the conversation Catalyst, Railway, and Axis are all having internally already.
The June 2026 HN thread is a one-month phase change. The May reading was that AI-native engineering hiring was a tooling preference. The June reading is that it's a gating criterion, and the JDs are now written in plain language so you can't miss it. Cursor Claude Code job descriptions are no longer a niche. They are the template.
The pool is not 4 people. It is somewhere north of 144,000. You just need a different index.
FAQ
Is the "bar is not coding" framing actually new, or did it exist in May?
May's HN thread had Cursor and Lovable as frequent mentions, but they read as preferences inside otherwise traditional JDs. AES's June post is the first time a serious hiring company has explicitly demoted raw coding to "not the bar" in the body of the JD, and Catalyst·Wayfare's "cover note matters more than a resume" reinforces it from a different angle. It is a one-month qualitative shift, not a continuation.
Should I still ask candidates for a resume?
Ask for both, but weight the artifact heavier. AES and Catalyst both treat the resume as table stakes and the artifact (CLAUDE.md, deployed app, cover note) as the real screen. If your ATS forces a resume upload, fine, but build the actual decision around the public work. Hiring managers reading the same JDs are doing this already.
How do I source for a role like AES's when LinkedIn returns 4 people?
Move the search off LinkedIn. GitHub code search for filename:CLAUDE.md, recency-filtered, gives you a real pool. Cross-reference with deployed-URL evidence (Vercel, Railway, Fly subdomain patterns) and public writing. Tools that index GitHub, LinkedIn, and the open web together in a single plain-English query, including Refolk, will compress that workflow from a half-day of tab-juggling to a single brief.
Does Boris Cherny saying "software engineer" might disappear actually matter for sourcing?
It matters as a directional signal more than a literal one. Titles change slowly, but the work has already changed. The practical implication is that you should stop using the title field as your primary filter and start using artifacts and outputs. Cherny's interview is the loudest version of a quieter truth the June HN thread is making explicit on the demand side.