Agentic Coding Hit 5% of HN "Who Is Hiring" Posts. Here's the Bullet.

In October 2026, 13 of the 261 company posts in Hacker News "Who is hiring?" named coding agents as a requirement for the role, up from 1 to 2% through May. For the first time, companies are hiring engineers who use Claude Code, Codex, and Cursor at scale, not engineers who build them. That means your resume needs a bullet it didn't need six months ago, and most engineers haven't written it yet.

The hook, in one chart

Agentic coding went from a 1 to 2% mention in HN hiring posts in early 2026 to 5% in October, which puts it on roughly the same trajectory "Git" had in 2012: differentiating now, table stakes soon. A scrape of twelve months of threads (solheimsolutions.no) shows a clean step up starting in June.

Within October, the Solheim table counts 5 posts matching "agentic coding," 3 matching "Claude Code," 12 matching "LLM," and 16 matching "AI agent." The agent language is now a distinct bucket from generic "LLM experience," and recruiters are using it as a filter.

Why most engineers' resumes are stuck on the wrong tool

The resume gap is backwards: Claude Code shows up 4.5 times more often than Cursor in Refolk's index of US professional profiles, even though Cursor is the more consumer-famous tool. In Refolk's index, 6,218 US professionals list "Claude Code" as a skill versus 1,381 for "Cursor." The mechanism is adoption velocity: Claude Code's workplace adoption rose from 18% in January 2026 to 39% by the May to July window (JetBrains data via uvik.net), and the resumes updated this year reflect that.

If your resume still names only Cursor, it reads as if you stopped updating it in late 2025.

The 4.5x skill-listing gap on resumes slightly overshoots the 3.25x usage gap in actual workplaces, which tells you something useful: resume language has already pivoted, so a Cursor-only line now reads older than the actual market is.

What hiring managers are actually asking for

They are not asking for "familiarity with LLMs." They are asking for a named agent, a named workflow primitive, and a throughput number. The language in October 2026 postings is explicit about ruling out dabblers.

A ZipRecruiter listing for "Gen AI Engineers (Claude code, Codex, Cursor)" in San Francisco at $144K to $190K asks for "Solid skills with Claude code, codex, Cursor" and the ability to "do large number of PRs per month." Another listing on the same board requires "Hands-on, real-world experience using Claude Code in development workflow, not just casual/occasional use." That phrase, "not just casual/occasional use," is the one you have to beat.

Google's October 2026 HN post for a Senior Staff SWE on the "CodeAI" team in Munich spells it out: they want someone "who has actually built agentic systems and has experience using them to make software development materially faster in a large engineering organization, not primarily a 'follow the space' role." Another October post in the same thread lists "strong TypeScript skills, deep practical knowledge of Git and worktrees, experience shipping production software, and hands-on experience with AI coding agents." Worktrees is the giveaway: that is the specific Git feature engineers use to run parallel Claude Code sessions on the same repo.

The named primitives you want to be fluent in, by October 2026:

• AGENTS.md and CLAUDE.md: repo-level config files that teach the agent your codebase conventions.
• Skills folders: reusable team-level instructions the agent loads per task.
• Hooks: PreToolUse, PostToolUse, and Stop hooks that gate the agent's behavior.
• MCP servers: Model Context Protocol servers that expose internal tools to the agent.
• Git worktrees: parallel checkouts so multiple agent sessions can run on the same repo without stepping on each other.

The resume bullet, by agent

Each major agent calls for a different proof point, because each has a different autonomy profile. Here is what to anchor on.

Claude Code

Claude Code bullets prove terminal fluency and in-loop iteration: you drive the agent from a terminal, course-correcting between tool calls. Name a workflow primitive and a throughput delta.

Example shape:

Shipped 18 to 22 PRs/week pairing Claude Code with a repo-level AGENTS.md and three PreToolUse hooks that gated DB migrations and external API calls; cut median review cycle from 2.3 days to 7 hours across a 400k-LOC TypeScript monorepo.

That bullet names the agent, names two primitives (AGENTS.md and PreToolUse hooks), and quantifies throughput against a before/after. A recruiter scanning for the "not casual/occasional" signal stops there.

Codex

Codex bullets prove spec-writing and eval discipline, because Codex runs async in a sandbox and returns a PR touching ten or more files hours later. You cannot course-correct mid-flight, so the scarce skill is writing the spec and the evals up front.

Authored 40+ Codex task specs with paired golden-output evals; multi-day runs averaged 11 files/PR with a 73% first-pass green rate on CI, replacing roughly one engineer-day of work per spec.

Cursor

Cursor bullets prove IDE-native collaboration and are still worth including, especially if the shop is a Cursor shop. But do not stop at Cursor if you have also used Claude Code or Codex.

Refactored a 60k-LOC Python service in Cursor using Composer with project-rules and a custom MCP server exposing our internal feature-flag API; shipped behind a flag in 4 days.

The anti-pattern

"Experience with AI tools including Cursor and ChatGPT." That reads, in October 2026, exactly like "proficient in Microsoft Office" read in 2015. It is the sentence the ZipRecruiter JD is written to filter out.

Writing three variants of a bullet, one per agent, and swapping them per posting is tedious, which is the exact work Refolk takes off you: paste the posting, get your own resume back rewritten against its specific agent and primitive requirements.

The sentence that beats "not casual/occasional use"

The sentence hiring managers are reading for is a judgment claim, not a tool list: what you decided to gate, what you decided to trust, and what you measured. An October 2026 HN post phrases the market shift precisely: "AI agents have made engineering throughput cheap, so senior judgment is now the scarce asset: what to build, what to gate, what to trust to production." (hnhiring.com/locations/remote)

Another October 2026 posting asks explicitly for engineers who "Ship spec-first" and "Orchestrate AI at staff level." Model your top bullet on that register. One format that works:

1. Verb of judgment: Gated, orchestrated, specced, evaluated, trusted.
2. Named agent and primitive: Claude Code with hooks, Codex with evals, Cursor with MCP.
3. Throughput or quality number: PRs/week, first-pass green rate, review-cycle time, incident rate.
4. Scope anchor: Repo size, team size, service criticality.

Where this leaves candidates in the next 90 days

At 5% of HN posts, you are early. Only 13 of 261 October postings named this, which means the bullet is still differentiating rather than table stakes, which is the window you want to be writing in.

Three moves worth making this week:

• Audit your current resume for the dated-tool tell. If "Cursor" is your only named agent and the posting named Claude Code or Codex, you will be screened out before a human reads it.
• Pick one primitive you have actually used and name it. AGENTS.md, hooks, worktrees, or an MCP server. One specific primitive beats five generic tool names.
• If you have no agent experience yet, take an evaluator contract. G2i is paying $100 to $200/hr for engineers evaluating "the quality of interactions with modern coding agents such as OpenAI Codex and Claude Code." That is a legitimate first-credential path.

The companies posting these roles right now are a specific list worth watching, because they are the ones who will accept an agent-fluent bullet instead of years of tenure:

• Google CodeAI (Munich): Senior Staff SWE, explicitly recruiting shippers of agentic systems.
• Scribd Engineering AI team: publicly claims the team "ships 100% of its software with AI coding agents like Claude Code."
• Lovable (Stockholm/London): recruiting for "the frontier of agentic coding."
• Zepto: self-describes as "AI-native," seeks engineers "already experimenting with AI and agentic tooling."
• Tether (AI Harness Engineer) and Axmed (AI Engineer, Agentic Systems): the new "agent harness" job family.

FAQ

Should I list Cursor if I mostly use Claude Code?

List both, but lead with Claude Code and anchor your strongest bullet there. In Refolk's index, Claude Code outnumbers Cursor 4.5 to 1 among US profiles, and workplace adoption data from JetBrains (39% vs 12% in mid-2026) tells the same story. Leading with Cursor signals you stopped updating your resume last year.

What if I have no production experience with coding agents yet?

Take a paid evaluator contract to generate a legitimate bullet in two to four weeks. G2i's listing pays $100 to $200/hr for evaluating interactions with Codex and Claude Code, and that work gives you a real credential: "Evaluated 300+ agent interactions across Codex and Claude Code sessions; authored rubric for tool-selection correctness." That is a stronger first bullet than a side project.

Is "agentic coding" the same thing as "building AI agents"?

No, and the distinction is now the whole point. HN postings in 2026 split into two families: engineers who build agent frameworks (Tether's AI Harness Engineer, Axmed's Agentic Systems role) and engineers who use agents to ship product faster (the Scribd and Google CodeAI roles). The 5% figure covers the second family, which did not meaningfully exist as a hiring category six months ago.

Will this bullet still matter in six months?

It will matter more, and it will stop being differentiating. The trajectory from 1% in January to 5% in October looks like the Git trajectory from 2010 to 2012: a specialist skill that became assumed. Writing the bullet now, while only 13 of 261 postings name it, is the window where it still moves you up a shortlist rather than keeping you from being cut from one.