By the Connecting Odds Research Team

Key takeaways

  • Nearly 4 in 10 software roles in our sample of live listings mention AI. The skill is no longer a separate specialty. It sits inside ordinary engineering work.
  • Senior software titles outnumber junior ones 16 to 1 in our sample. The ladder did not disappear, but the bottom rung got much narrower.
  • Developers use AI tools every day and mostly do not trust what they produce. That gap, between daily use and low trust, is the new job skill: checking the work.

Three years ago, an AI coding tool finished your line of code. Today it plans a change across several files, writes it, runs the tests, and hands you the result to check. The job title stayed the same. The work did not.

This article looks at how software work actually changed, and what the hiring market says about it. We sampled 5,001 live listings from the Connecting Odds public jobs feed on November 13, 2026, and counted what employers ask for. The numbers are ours. The survey figures come from JetBrains and Stack Overflow, and we name them where we use them.

The three rewrites

The change did not arrive as one big event. It arrived as three smaller ones, and each one moved a different part of the job.

First, autocomplete became an agent. The early tools suggested the next few words while you typed. The current tools take a plain-language request, read the codebase, make a plan, change several files, and run the tests. You describe the outcome. The tool does the typing. That is a real change in who does what, and it is why the JetBrains 2026 Developer Ecosystem Survey found 90 percent of professional developers using AI coding agents at work at least weekly, with 68 percent using them daily. Those figures come from a survey of more than 15,000 professional developers, run May to July 2026.

Second, writing code became reviewing code. When a tool can produce a working first draft in a minute, the scarce skill stops being typing speed and starts being judgment. The developer reads the diff, checks the logic, spots the invented API call, and decides whether the change is safe. This is the same prompt, review, verify workflow we covered in the second article in this series, applied to code instead of text. The review step is now the job.

Third, the tool choice stopped being the point. In 2024, teams argued about which assistant to buy. In 2026, the tools change every few months, and the workflow matters more than the brand. The developers who do well are not the ones who picked the right tool in January. They are the ones who learned to brief a tool clearly, check its output, and keep working when the tool changes underneath them.

What the listings say

Here is what our own data shows. We pulled 5,001 live listings from the public jobs feed and looked for AI in the title, the summary, or the listed skills.

Almost a third of all listings, 30.9 percent, mention AI somewhere. That is not a third of tech jobs. It is a third of everything in the sample, across all kinds of roles. Narrow it to software roles, the 1,229 listings for engineers, developers, and related titles, and the share rises to 38.2 percent. Nearly 4 in 10 software roles now mention AI as part of the work.

The skills tell the same story from a different angle. The single most common skill tag across the whole sample is artificial intelligence, attached to 1,173 listings, or 23.5 percent of everything we pulled. Machine learning follows at 416, then generative AI at 185 and large language models at 167. Python, the language most associated with AI work, is the third most common skill tag overall, behind only artificial intelligence and Excel.

Only 201 listings, 4.0 percent, actually carry AI in the job title. The rest are ordinary software roles where AI is now part of the description: a backend engineer who uses AI-assisted tooling, a data engineer who works with model pipelines, a product engineer whose product has AI features. The job is not called an AI job. The work still includes AI.

One detail surprised us. Not a single listing in our sample names a specific AI coding tool: no Copilot, no ChatGPT, no Cursor, no Claude, no Gemini, no Codex, no Devin. Employers ask for AI skills in general terms, not for experience with one named product. That matches what we found in the first article in this series, our plain guide to what each tool does. The tools change too fast for a job posting to bet on one. What employers want is someone who can work with AI tools in general, and that is a skill you can build without waiting for a posting to name the tool.

The senior-junior split

The most striking numbers in our sample are not about AI at all. They are about seniority.

Across all 5,001 listings, 44.9 percent carry a senior title: senior, staff, principal, lead, manager, director, head of, or architect. Only 4.7 percent carry a junior or entry-level title. That is roughly 9 senior titles for every junior one.

Software is more lopsided still. Among the 1,229 software roles, 53.0 percent are senior and 3.3 percent are junior or entry level. That is 16 senior software titles for every junior one. And the AI-flavoured software roles skew senior too: 61.7 percent of software roles that mention AI carry a senior title.

We should be careful about what this proves. Job boards have always listed more senior roles than junior ones, because junior roles are often filled through internships, campus hiring, and referrals rather than public postings. Our sample is public listings, not all hiring. Even so, a 16 to 1 ratio is a warning sign for anyone starting out. The work that used to train junior developers, boilerplate code, simple tests, routine fixes, is exactly the work AI tools now do first. The bottom rung did not vanish, but it got narrower, and the way up changed: juniors now have to show judgment, not just output, earlier than before.

For hiring teams, the same numbers point the other way. If most of the public market is senior, then senior engineers are the ones competing openly, and junior talent is mostly found elsewhere. Posting a junior role and waiting for applications is a slow way to hire. The teams that hire juniors well will keep doing it through internships, apprenticeships, and structured onboarding, not through a listing.

The trust gap

The last piece of the picture comes from the developers themselves. The Stack Overflow 2025 Developer Survey, fielded May to June 2025 with more than 49,000 developers, found that 84 percent were using or planning to use AI tools, and about half of professional developers used them daily. But only about a third said they trust the accuracy of AI output, and just 3 percent said they highly trust it. More developers distrust the output, 45.7 percent, than trust it.

Put the two surveys together and the shape of the new job is clear. Nearly everyone uses the tools. Nearly everyone checks them. Daily use and low trust live side by side, and the work that sits between them, reading, testing, and taking responsibility for machine-written code, is the skill employers are actually paying for. That is why the senior roles dominate the listings. Reviewing is a senior skill, and the market is hiring for it even when the posting never says so.

What to do about it

If you write software for a living, three things follow from all of this.

Learn the review, not just the tool. The tool you use today will be different next year. The habit of reading a diff carefully, running the tests, and refusing to approve code you do not understand will still be the job. Practice it on every AI suggestion you accept, even the small ones.

Build proof of judgment, not proof of output. A portfolio that shows ten projects written with AI help says little. A portfolio that shows what you caught, what you fixed, and why you rejected a plausible-looking solution says much more. When you apply, our resumes playbook has the detail on presenting real work: Resumes and Job Search: The 2026 Playbook.

If you are starting out, aim at the work AI does badly. Messy debugging, integrating systems that were never designed to work together, talking to users to find out what they actually need, and reviewing other people's code are all hard to automate. Those are also the tasks that teach you the most. The junior playbook for this new market is the subject of a later article in this series, and the short version is this: get close to real systems and real users as fast as you can.

If you are hiring, the listings data has a plain message too. AI skill is now a normal part of a software role, so asking for it separately adds little. Ask instead how a candidate checks AI output, and give them a real piece of machine-written code to review in the interview. You will learn more in twenty minutes than a skills list will tell you in a week.

How we counted

Our sample is the first 5,001 listings returned by the Connecting Odds public jobs feed on November 13, 2026. The feed re-orders as listings are re-verified, so a different pull on a different day would give slightly different shares. The sample is a snapshot, not a census.

A listing counts as mentioning AI if the title, summary, or skills contain artificial intelligence, machine learning, generative AI, large language models, natural language processing, computer vision, prompt engineering, or a related term. That is a broad net, so 30.9 percent is an upper measure of AI touching the role, not a count of AI jobs. Only 4.0 percent of listings carry AI in the title.

Senior means the title contains senior, staff, principal, lead, manager, director, head of, or architect. Junior means it contains junior, entry level, graduate, intern, or associate. Titles with neither are counted as mid or unspecified, and they are the largest group overall at 48.9 percent. All 5,001 listings in this sample were imported from employer career pages, the same pattern we found in our earlier look at the hidden job market: The Hidden Job Market, Part 2.

The survey figures are JetBrains 2026 Developer Ecosystem Survey, May to July 2026, more than 15,000 professional developers; and Stack Overflow 2025 Developer Survey, May to June 2025, more than 49,000 developers. Both were re-checked on the day we drafted this article.

Frequently asked questions

Has AI replaced software developers? No. It replaced a slice of the work, mostly routine writing and boilerplate, and it changed the rest. Developers now spend more of their time reviewing, integrating, and deciding, and less time typing first drafts. The listings back that up: senior titles dominate, and AI shows up inside ordinary roles rather than replacing them.

Do I need to learn AI tools to get a software job? Nearly 4 in 10 software roles in our sample mention AI, and 90 percent of professional developers use AI coding agents at least weekly, according to JetBrains. You do not need to master every tool, but you do need to be able to work with one, and to check what it produces. Our plain guide to what each tool does is a good place to start: AI Tools in 2026: A Plain Guide to What Each One Does.

Why are there so few junior developer roles? Our sample shows 16 senior software titles for every junior one, but public listings are only part of hiring. Junior roles are often filled through internships, campus programs, and referrals, so they appear less in public feeds. The AI shift still matters: the routine tasks that used to train juniors are now the first tasks tools take over, so juniors need to show judgment earlier.

Should I trust AI-written code? Trust it the way you would trust a draft from a fast but careless colleague: useful, often right, and never to be accepted without reading it. Only 3 percent of developers in the Stack Overflow 2025 survey highly trust AI output. The working method is prompt, review, verify, which we covered in detail here: Using AI Tools Well: The Prompt, Review, Verify Workflow.

What AI skills should I put on my resume? Put the ones you can actually demonstrate, and name what you did with them, not just the term. A line that says you reviewed and tested AI-generated code for a real project beats a list of tool names. And never let a tool invent experience for you: a hallucinated skill on a resume ends a candidacy.

Next in this series: The New Rules: Usage Billing, Caps, and AI Tool Policies, on how the December 2026 shift to usage billing changes what teams and freelancers pay for.