About one in six do, in writing, and the ones that do ask for more than the tool. Of 416 software engineering postings we collected on 29 September 2026, 67 (16%) named an AI coding tool or asked for AI-assisted development in so many words. Claude Code was the tool named most (27 ads), then Cursor (about two dozen), Codex (12) and GitHub Copilot (10). The other five in six say nothing either way, which in 2026 mostly means the team already assumes it.
What matters more than the headline is what those 67 ads ask for alongside the tool.
Which AI coding tools do job ads name?
Our sample is the same public fetch behind our count of AI job ads, re-run on 29 September: Hacker News "Who is hiring?" threads for July to September 2026, Remotive and Arbeitnow, 2,610 postings in all. A posting counted as software engineering when its title said software engineer or developer, full-stack, back-end or front-end. Among the 416:
| Named in the ad | Ads |
|---|---|
| Claude Code | 27 |
| Cursor | about 24 |
| Codex | 12 |
| GitHub Copilot | 10 |
| "AI-assisted development", "coding agents" or "AI coding tools", no product named | 42 |
| Any of the above | 67 (16%) |
Most ads that name a tool name several and add "your choice". The Cursor count is approximate because the word also appears as a company name and in one ad that bans it outright ("No Cursor."). The counts file is public, with the script beside it.
What else do those ads ask for?
Judgement, mostly. Among the 67 ads that name AI coding tools:
- Production or shipped systems: 49%.
- System design or architecture: 48%.
- Ownership or autonomy: 48%.
- Testing: 30%.
- TypeScript or JavaScript: 51%, and Python 37%.
Read together, the ads describe the same person: someone who uses agents to write more code, and who owns what ships. The phrasing is telling. One ad wants a candidate who builds with AI IDEs "as a force multiplier", not occasionally; others ask you to review, test and take responsibility for agent-written code. None of the 67 treats the tool as the skill.
Will you fail an interview for not using AI coding tools?
At some companies, now, yes; at others you will fail for leaning on them. Interviews are splitting three ways:
- AI allowed and watched. You build a small feature with the tools you normally use while an interviewer watches how you prompt, read and correct the output. Using no tool at all reads as slow.
- AI banned. A live exercise with no assistant, to check you can write and debug code yourself. The "No Cursor." ad is this camp.
- Take-home, anything goes, then a deep review. The code is the starting point; the interview is you explaining every line and changing it on the spot. This is where people who let an agent write things they do not understand come unstuck.
The safe preparation covers all three: be fluent with one agent, and be able to work without it.
What does "good with AI coding tools" actually mean to an employer?
Not typing speed. Four habits separate the engineers employers describe from the ones they are wary of:
- Tests as the contract. You write or edit the test first and let the agent satisfy it, so correctness is defined by you, not by the model.
- Reading every diff. You can explain each change the agent made, in your own words, and you reject the ones you cannot.
- Context engineering. Project instructions (a CLAUDE.md or equivalent), small tasks and clear plans, because an agent is only as good as what it is told.
- Knowing when to stop. Security-sensitive code, migrations and anything you cannot verify get written or checked by hand.
A useful self-test: take a feature an agent helped you build and delete it. Could you rebuild it without the agent in a day? If not, you are depending on it rather than directing it.
Is it still worth learning to code the hard way?
Yes, because the review is the job. An agent writes a plausible pull request in a minute; knowing whether it is correct, secure and maintainable still takes an engineer. Even the ads that name a tool ask for production experience, system design and ownership more often than they name any single tool. The tools make a good engineer faster and make a weak engineer's mistakes arrive faster too.
For someone learning now, the sensible order is: learn to program and debug without help first, then add an agent and learn to direct it, then learn to review its work as if a junior engineer wrote it.
How do you show an employer you can do it?
Deployed work with the evidence attached. A repository where the tests came first, the commit history shows your review, and the README says what the agent did and what you decided. One project that fixed a defect and proved it with a failing test that now passes is worth more than ten generated apps.
Square 1's courses are built around exactly that record. AI Coding Workflows is an on-demand course that builds one feature three ways, in Cursor, Claude Code and Codex, and teaches when to reach for which. Claude Code from Zero sets up a daily Claude Code workflow on a real codebase. The Software Engineering with AI Bootcamp is twelve weeks, live on Zoom with one instructor, about 15 hours a week: a full-stack product built with coding agents, tests and CI, a squad build through code review, a hardening pass where every fix is proved with a test, and a recorded viva defending the code you and the agent wrote. The full-stack engineer role page lists the day-to-day, and the free full-stack skill check takes about three minutes.