One Engineering Posting in Five Now Has an Opinion About Your AI Tools
Elena VasquezHead of Talent
Last updated:
September 15, 2026
We read 4,820 software engineering postings from 179 tech companies. One in five now has a clause about using AI coding tools, one in ten makes it a requirement, and Cursor is named more than Copilot.
Table of content:
What 4,820 software engineering postings say about the engineer using AI coding tools, September 2026
What the posting says
Postings
Share
Required: listed among the qualifications
487
10.1%
Preferred: an explicit plus
74
1.5%
Expected: stated as how the team works
359
7.4%
Provided: the company pays for the tools
43
0.9%
Any of the above
963
20.0%
No clause of this kind
3,857
80.0%
Read on 15 September 2026 from the public Greenhouse, Ashby and Lever job boards of 179 technology companies. A posting takes its strongest label, so the 43 provided rows are the ones that say nothing stronger; 77 postings mention provided tools in all. Every posting, the sentences behind its label and the labeling rules are published in the full dataset.
“Reach for Claude Code, Cursor, Copilot, or whatever helps, and be clear about where you did and didn’t.”
That sentence sits under “About you” in a ClickHouse posting for a senior engineer. It is a qualification, printed in the same list as Go and Kubernetes, and a year ago it would have been an odd thing to find there.
So we went looking for how common it has become. We read the public job boards of 179 technology companies, kept every software engineering posting, 4,820 in all, and pulled out each sentence that mentions an AI coding tool or AI-assisted development. Then we sorted those sentences by what they actually ask of the person applying. Every posting and every sentence is published, so the counts below can be argued with row by row.
One in five overall, one in four since July
963 of the 4,820 postings, 20 percent, say something to the candidate about using AI coding tools. Not building AI products, which is a different and much larger category and a different hire. Using Cursor or Claude Code or Copilot to do the ordinary job. Half of those 963 make it a requirement; the table above has the split.
The number that matters is the slope. Every posting carries the date its board published it, so a snapshot taken on one day can be read, with care, as a record of what companies were writing when. Among postings published in January to March 2026, 11 percent had a clause of this kind and 5 percent made it a requirement. Among postings published in July to September, 26 percent and 14 percent.
Postings with any clause about the engineer using AI coding tools, by the month the board published them, with the number of postings behind each bar. September is a partial month. Source: the 4,820-posting dataset linked above.
Two cautions before reading that as a trend line. The rise is not monotonic; it falls in three of the eight steps, and January rests on eight postings. And the boards allow edits in place, so a posting can carry a January date and July text, while a January posting still open in September is an unusual posting to begin with. Those two biases pull in opposite directions, and neither is measurable from the outside. What can be checked is whether the gap is an artifact of a few boards. It is not: among the 27 companies with at least five postings published in both the first quarter and the third, the share with a clause went from 8 percent to 23 percent, and it rose at 18 of the 27. Averaging companies instead of postings, so that no single board carries the number, gives 13 percent for the first quarter and 27 for the third.
What “required” actually says
Skip the screening, get vetted candidates in 48h
Senior engineers at $50 to $80/hr, each through a paid 10-hour assessment. No fee until you hire.
At the soft end, Asana’s postings ask for “curiosity about AI tools and emerging technologies, with a willingness to learn.” At the hard end, an Okta posting lists, under Requirements and without a verb to soften it, “Daily use of Claude Code, Cursor, GitHub Copilot, or equivalent.” Samsara wants someone who has “shipped real work with AI coding tools (Claude Code, Cursor, Copilot, custom agents), not just tried them out.”
What the strict ones have in common is a rider. Nubank asks for fluency “with good judgment about where they help, where they don’t, and how to review their output critically.” Pinterest wants a “strong track record of critical evaluation and verification of AI-assisted work.” Stripe’s line is the most quotable: “fluent with AI-assisted development but not dependent on it; you can reason through problems and debug without AI as a crutch.” Twilio simply says “you should be able to describe your workflows and show examples.”
Read together, the requirement is “use AI and be able to say what you did with it.” Most candidates are rehearsing the first half.
Cursor and Claude Code first, Copilot third
270 postings name at least one tool, and the list is short. Cursor appears in 135 postings from 50 companies and Claude Code in 116 from 49; a one-company gap is a tie. Copilot follows in 90 postings from 30 companies, counting the 46 postings that write just “Copilot” without GitHub in front, then OpenAI’s Codex agent in 76 from 23. After that the list thins out fast: Windsurf at two companies, and Cline, OpenCode, Devin and JetBrains’ Junie at one or two each.
Companies naming each tool in a sentence addressed to the candidate. Mentions of a company's own product, or of its customers, are not counted. Source: the same dataset.
The incumbent is third. Copilot went on general sale in June 2022, with a home inside GitHub; Cursor shipped in 2023 and Claude Code in early 2025, and both now outrank it. Seventy-four postings name Cursor without mentioning Copilot at all; 21 name Copilot without naming Cursor or Claude Code. Most lists end in “or similar,” so these are examples rather than mandates, but examples are what a hiring manager reaches for first.
Who is asking, and it is not the AI labs
The companies that build these tools mostly do not put the clause in their postings. The sample holds 30 AI labs and AI tool makers, from OpenAI and Anthropic to Cursor itself, with 1,026 postings between them; 4 percent of those postings make AI tool use a requirement. The other 149 companies, with 3,794 postings, are at 12 percent. OpenAI and Anthropic postings talk about Codex and Claude constantly, but as the product; the sentence asking you to use them shows up at a payments company, an HR platform, an identity vendor.
The postings that carry it most often come from Pinterest, Okta, Affirm, Block, Fivetran, Stripe, Toast and Asana. Anthropic is on the list too, with 14 of its 150 postings, a below-average rate for a board that size. 101 of the 179 companies have at least one posting that requires it; 128 have at least one that mentions it.
It also climbs with seniority, once Cloudflare is set aside, because its company description carries the same line in all 60 of its postings.
Postings that list AI coding tool use among the qualifications, by the seniority in the title. Cloudflare excluded. Intern n=217, mid n=1,673, senior n=1,619, staff n=1,228. Source: the same dataset.
Staff and principal postings ask for it more often than mid-level ones, and interns and new graduates are asked as often as staff. The two ends of the ladder are being screened for the same thing.
487 postings ask you to bring the skill. 77 say they will pay for the tool
The 77 come from eight companies, and they are specific. Ashby offers “$250/month tool budget plus access to Claude, Cursor, and Codex with generous token budgets.” GitLab says it provides Claude Code. Grafana Labs promises “access to frontier models” and a company-funded usage budget. Samsara mentions “generous token allotments.”
Everyone else is silent on who pays, which in practice means the candidate is expected to have learned on their own subscription, on their own time.
The interview paradox
Only ten of the 179 companies say anything about AI in their own hiring process, and they disagree.
Peloton does not permit “the use of AI tools during any stage of the application and interview process.” Two of its eight postings, meanwhile, expect the hire to lean on “AI-first coding tools (like Claude Code)” from day one. Toast has an AI clause in 45 of its 47 postings, twenty of them as a requirement, and asks candidates to “complete interviews and assessments on your own, without real-time AI tools.” 1Password welcomes AI in the application and asks you to leave it at the door of the live interview.
On the other side, Customer.io runs a “Practical AI-Enabled Coding Round,” a live hour where you extend a sample SDK with AI tools. Zapier publishes a guide on how to collaborate with AI during its hiring process. Coder permits AI on take-home assessments and says so.
There is a defensible case for the first camp: an interview without the tools is one way to test the judgment the riders ask for, and the tools come back on day one. But almost nobody makes that case in writing. Companies want the skill, have not agreed on how to test it, and mostly have not said.
What to do with this
If you are the candidate, the clause is a screening question, and the riders tell you the shape of the answer. Have one story ready: a real task, what you reached for, what it got wrong, how you caught it, what you would not hand to it. And check the interview rules before you walk in; ten companies wrote them down, and the other 169 will tell you in the room.
If you are the one hiring, a bullet that says “daily use of Claude Code” screens for people who can write that bullet. What the paradox companies are groping toward is a better test, and it is not hard to run: a scoped paid task of a few hours, followed by a short writeup of what the tool got wrong and what the person changed. The writeup, not the diff, is the signal, and no interview question gets at it as cheaply.
What this sample is and is not
It is 179 technology companies whose job boards run on Greenhouse, Ashby or Lever and answer a public request, read on 15 September 2026. That skews toward US tech and toward companies large enough to run a board at all. It is postings, not practice: a clause is what a company writes, not what its engineers do on a Tuesday.
Sentences were pulled by rules and then labeled one by one by a language model against written instructions, which are published with the data; an independent second pass over a random 100 of them agreed with 98, and neither disagreement touched a “required” label. One Cloudflare line that repeats in all 60 of its postings was counted as an expectation rather than a requirement, and the reasoning is in the appendix. A title with no seniority marker counts as mid, which makes that bucket the noisiest. The month figures use each board’s own publication field, and boards allow edits in place, so they mix vintage with survival and with revision.
This website uses its own and third-party cookies for analytical purposes and to offer you advertising of interest to us and third parties. You can consult all the information in our Cookies Policy. You can manage the acceptance or rejection of cookies by clicking on “Configure”.
Privacy is important to us, so you have the option of disabling certain types of storage that may not be necessary for the basic functioning of the website. Blocking categories may impact your experience on the website. More information
