Which AI tools do employers ask for?

Engineering job posts have long listed the tools of the trade: Git, code review, a test suite. In 2026 a new line is turning up next to them. Some employers write that they use Claude Code or Cursor every day, some say candidates should already be using them, and a few say the opposite. Nobody publishes how often this happens, or which tools get named, so we counted.

We read every job post in Hacker News's monthly "Ask HN: Who is hiring?" thread, from April 2011 to October 2026: 187 threads and 95,098 posts. HN's board is mostly startups, so on 31 October 2026 we also downloaded every live engineering posting from the public job boards of 51 tech employers, and set both next to Indeed's postings data and what developers say they use. The methodology explains how, and where the counts fall short.

AI coding went from a footnote to a line item#

Before 2025 almost no HN job post talked about how the team writes code with AI. We matched tool names (Cursor, Claude Code, GitHub Copilot, Codex, Windsurf and a dozen smaller ones) and practice phrases ("AI-assisted development", "coding agents", "agentic coding", "vibe coding", "AI fluency"). We removed each company's own name first, so Cursor's own job posts don't count as mentions of Cursor.

The yearly shares tell the same story with less noise: 0.75% in 2023, 0.51% in 2024, 2.86% in 2025 and 8.69% in 2026. The 2023 bump isn't what it looks like. Of the 12 companies flagged that year, 9 were building a coding assistant themselves (Sourcegraph, TabbyML, Replit and others) and the other 3 used the words in another sense, such as a founder's bio. None said its own engineers used one.

The first post we found that named an AI tool as part of a team's own work came in October 2024, in a list of the tools the team uses. The first that asked for it came in December 2024, for an expert user of an AI-assisted IDE. Claude Code's first mention was in July 2025.

Counted by company instead of by post (companies repost every month), the numbers are slightly higher: 11.4% of the companies hiring on HN in 2026 mention AI coding, and 6.2% name a tool.

Mentioning is not requiring#

A regex can tell you a post says "Cursor". It can't tell you whether the company wants you to use it, or is pitching itself as "Cursor for radiologists". So we read the 2026 post of every company the pattern flagged, 192 of them, and coded each one by hand.

The requirements are often blunt. An Android role at Jefit calls fluency with AI coding assistants non-negotiable. Search Atlas wants people who already use Claude Code or similar every day. Veedma asks for six months of daily Claude Code use, including writing skills and wiring up MCP servers. Hospitable says 90–95% of its production code is AI-generated, with Cursor and similar tools. A summer internship at AES says Claude Code writes all its code, so the bar is something other than raw coding ability.

Others set limits. Kinxshn uses AI tools heavily but says it doesn't vibe code. Stpkr Technologies wants people who can do the work without AI and do it much faster with it. Splash Tech, the one company coded as discouraging it, allows AI assistance but rules out Cursor.

Entry-level roles aren't exempt. 9.6% of 2026 junior and intern posts mention AI coding (n = 167), against 8.4% of senior posts (n = 1,355), which is no meaningful difference at these sizes. Posts with AI in the header line (the company name, a .ai domain or the role title) mention it more often (12.5%) than the rest (7.8%).

Claude Code is the tool employers name#

Counting every post that names each tool (a post can name several), Claude Code is ahead in 2026.

HN is one board. The employer job boards give a second view, and there the two leaders are level: Cursor in 103 postings, Claude Code in 99, Codex in 69 and GitHub Copilot in 66, out of 2,620.

That's behind what developers say they use. JetBrains surveyed professional developers in May to July 2026: 39% used Claude Code at work, up from 18% in January, and 47% in the US. GitHub Copilot fell to 21% from 29% a year earlier, and Cursor to 12% from 18% in January. Codex went from 3% to 16%.

Job descriptions are often reused for months, which may be why they trail usage. Either way, a posting that names one tool may say more about when it was written than about what the team uses today.

AI is now in most HN job posts#

The AI-coding line is the newest layer of a larger shift. In 2026, 55% of HN job posts use the word "AI" or a term like LLM, GPT, generative AI, OpenAI, Anthropic or Claude. In 2022 it was 8.6%.

The vocabulary changed with it. "Machine learning" and its relatives (deep learning, neural nets, PyTorch, TensorFlow) peaked at 13.3% of posts in 2018 and are down to 6.9%, their 2012 level. In their place, 20.5% of 2026 posts mention building with LLMs (agents, RAG, embeddings, fine-tuning, evals, MCP), against 0.5% in 2022. "Agents" or "agentic" alone appear in 16.3%.

The board shrank, but not its AI posts#

HN's hiring thread is much smaller than it was. It averaged 857 job posts a month in 2021 and 291 in 2026, and the August 2026 thread had 222, the fewest since January 2014. The fall is entirely in posts that don't mention AI or machine learning.

It's tempting to read this as AI jobs replacing the rest. The data can't show that. Many posts count as "AI" because the company now describes its product that way, or because of a line about using AI tools, and the same companies may be hiring for the same roles they always did. The board's size also depends on HN itself: its audience, its moderation and the other places startups now post. What it does show is that the HN thread lost most of its non-AI posts after 2022, the same years the wider US software postings collapse happened. Indeed's US software development postings index stood at 80.8 on 23 October 2026 (1 February 2020 = 100), against 103.7 for all postings.

Outside HN, one engineering posting in four#

To check HN against a different set of employers, we took a fixed list of well-known US tech companies that publish their job boards on Greenhouse, Ashby or Lever, and downloaded every live posting on 31 October 2026. We kept engineering titles, removed duplicates, and left out companies that make AI models or coding tools (they name their own products). That left 2,620 postings at 51 employers, from Airbnb and Cloudflare to Stripe and Waymo.

So the broad share is 23.4% (613 postings), the strict one 16.2%, and only 6.4% name a tool. Where a tool is named, it's usually several: 111 of the 168 postings that name one name two or more.

Posting shares can mislead here, because many employers paste one AI line into every posting. GitLab, for example, expects all team members to bring AI into their daily work, so all 64 of its engineering postings say so. The employer view is steadier.

The same postings show which AI skills employers want. Agents lead, and prompt engineering is near the bottom.

The wider market is moving the same way#

Large posting datasets don't publish counts by tool, but they show AI language spreading across all jobs. Indeed's AI Tracker measures the share of postings that mention AI, machine learning or data science. In the US it was 7.15% in September 2026, twice the 3.56% of September 2025 and about four times the 2019 average.

Other datasets agree on direction, if not on level, because each uses its own keyword list. Lightcast's data for the Stanford AI Index found that about 2.5% of US job postings mentioned AI skills in 2025, and that agentic AI skills went from 0.06% to 0.23% of postings in a year. Indeed's economists found US software development postings up almost 15% since Claude Code's launch while postings overall fell 7%, and that 71% of the increase between May 2025 and May 2026 came from senior roles.

Mandates, interviews and pay#

The job-post lines follow a year of memos. Shopify's chief executive told staff in April 2025 that using AI was now a baseline expectation, that it would count in performance reviews, and that teams should show AI can't do a job before asking for headcount. Others followed.

Interviews are split. Canva replaced its computer-science fundamentals screen with an AI-assisted coding interview, on the grounds that assistants can solve the old questions. Anthropic, which makes Claude, asks candidates to write their own first draft of an application and to do live interviews without AI unless told otherwise.

Some developers see it the same way. In "Developers, reinvented", an August 2025 post by Thomas Dohmke based on interviews with 22 developers, one of them put it like this:

Whether it pays is less clear. PwC's 2026 AI Jobs Barometer, built on more than a billion job ads in 27 countries, puts the average wage premium for AI skills at 62%, up from 57%. That compares the advertised wages of postings that ask for AI skills with those that don't, within a sector; it isn't a raise for learning a tool. On the employer side, Payscale's 2026 survey of employers (3,413 responses) found 55% offer no premium, bonus or equity for AI skills, while 31% said they were rewriting job descriptions for IT roles to require AI competencies.

What this means if you build with AI#

1. Put the tools and the workflow on your CV. Nearly one in ten HN job posts this year and one in four postings at the employers we checked mention AI tools. "Built the billing service with Claude Code, with tests and review" says more than "AI-savvy". Name the tool, and say what you did with it.
2. Be fluent in the category, not one product. When postings name a tool, they usually name two or more. Usage moves fast: Cursor fell from 18% to 12% of developers in six months while Claude Code rose from 18% to 39%. Show that you can pick up the next one.
3. Learn the skills employers actually list. Agents, LLM integration, MCP and evals appear far more often than prompt engineering. A small agent with an eval suite is a better portfolio piece than a list of prompts.
4. Be ready to work without it. Interview rules differ: Canva expects AI in coding interviews, Anthropic forbids it in live ones unless stated. Ask the recruiter which, and practise both ways.
5. Write your own application. 15 HN posts in 2026 say they will discard AI-written applications or messages. An AI-generated cover letter for a job that wants AI skills is still a cover letter nobody wanted.
6. Hiring managers: say what you mean. "Uses", "encourages" and "requires" are different policies, and fewer than half of the HN companies that mention AI coding require it (74 of 192). State the rule for the interview too.
7. Name the outcome, not last quarter's tool. A posting that lists one product dates quickly. "Ships with AI coding agents and reviews their output" will still be true next year.
8. Don't promise a premium you don't pay. The headline premiums compare postings, not people. If AI fluency is now the baseline, say so, and pay for the outcomes.

We build host0 for the step after your agent writes the code: what it made goes live at a private link you can share by email. A portfolio piece built with Claude Code is easier to judge when the hiring manager can open it.

Methodology#

All data is public and was collected on 31 October 2026. Our own counts are reproducible from the scripts in our repository.

• HN threads. Every monthly "Ask HN: Who is hiring?" thread from April 2011 to October 2026 (187 threads), from the HN Algolia API, cross-checked against the HN Firebase API so no live post was lost. A job post is a live top-level comment; 99 job-seeker posts misfiled in the thread are excluded, leaving 95,098. 2011 runs from April and 2026 to October, so yearly figures are shares or per-thread averages.
• Flags. Regexes on each post's text with URLs and emails removed. AI-coding mentions are tool names (Cursor, Claude Code, GitHub Copilot, Codex, Windsurf and smaller tools) or practice phrases, with the posting company's own name removed first. "AI" means the word AI or artificial intelligence, or LLM terms such as GPT, ChatGPT, generative AI, OpenAI, Anthropic, Claude and Gemini. In hand checks of random samples, every tool flag was right in all of its 13 to 34 sampled hits.
• Hand-coded census. We read the post of every company flagged for AI coding in 2023–2026 (290 companies; 192 in 2026) and coded it as requires or expects, uses or encourages, builds an AI coding product, other sense, or discourages. One coder, not blind; treat "requires" as ±5 companies.
• Recall. In 80 random unflagged 2026 posts, 2 described AI-assisted coding in words the regex missed. So the pattern finds roughly three in four such posts, and 8.7% is likely an undercount of mentions. Of the flagged 2026 companies, 78% really describe using or expecting AI coding tools.
• Employer snapshot. Public job-board JSON (Greenhouse, Ashby, Lever) for a fixed list of 83 job-board accounts of well-known tech employers, chosen before looking at results and downloaded once on 31 October 2026. Engineering titles only, deduplicated by company, title and text, AI model and coding-tool makers excluded: 2,620 postings at 51 employers. It's one day, not a trend; not a random sample; it includes non-US roles; and it misses Google, Meta, Microsoft, Amazon, Apple and Shopify, which run their own job sites. Several employers paste the same AI line into every posting, so read the employer median (18.9%) next to the posting share. Hand checks: 88% of strict matches and 93% of broad-only matches were right.
• Indeed. Indeed Hiring Lab's AI Tracker (October 2026 release, data to 30 September 2026) and its postings indexes on FRED (to 23 October 2026). Indeed revises past values by about 0.1 points between releases. We leave Indeed's Canada series out: it tripled in nine months while every other country rose about 1.5×, which we can't explain.
• Other sources. JetBrains, PwC, Payscale, Lightcast and company statements are their own numbers, linked where they're cited. JetBrains and PwC were read from archived copies where the live page blocked us.

Limits. HN's hiring thread is one community's board, skewed to startups, the US and Europe, and engineering roles. It is a culture indicator, not a labour-market statistic. Every share except the hand-coded census counts mentions, not requirements. Monthly tool counts are single digits to mid-teens, so we only compare tools over several months.

The data pack, with every source, tier and a do-not-cite list, is in our repository at docs/research/2026-10-31-ai-in-job-posts.md. The series behind every chart and the scripts that compute them are in docs/research/2026-10-31-ai-in-job-posts-data/.

Open questions#

• Do companies that require AI coding tools hire fewer juniors? We found no difference in how often junior and senior posts mention AI coding, but 167 junior posts in 2026 is too few to answer it.
• How many developers work under a mandate? Neither the Stack Overflow nor the JetBrains survey asks whether an employer requires AI tools. JetBrains' 2025 survey found 68% of developers expect it to happen.
• What do Big Tech postings say? Google, Meta, Microsoft and Amazon aren't in our snapshot, and they employ a large share of the engineers who use these tools.
• Does the AI wage premium survive controls? PwC and Lightcast compare averages across postings. Nobody has published a premium controlled for seniority and employer.
• Will the share keep climbing? Pooled over August to October 2026, one HN post in ten mentioned AI coding. The November and December threads will show whether that's a plateau.