The developer’s job in 2026
Coding agents now write much of the code, juniors find fewer doors open, and the productivity studies disagree. What changed, what the evidence says, which skills matter now.
Two years ago an AI assistant finished a developer’s line. Today a developer describes a task, an agent reads the codebase, writes the change, runs the tests and proposes the result, and the developer decides whether it is good enough. The job has not disappeared. It has moved. Here is what changed, what the studies say about speed, what is happening to hiring, and which skills matter now.
What changed
The tools went from optional to normal within a year. In Stack Overflow’s 2025 survey, 84 percent of developers used or planned to use AI tools, but only about three in ten used agents at work, and the survey called agents “not yet mainstream”. In the 2026 edition, with more than 30'000 respondents, 73 percent said they use coding assistants and agents every day. Claude Code and GitHub Copilot lead among the agents.
Trust has not kept pace. In 2025 more developers distrusted the accuracy of AI tools, 46 percent, than trusted it, 33 percent, and the most common frustration, named by 66 percent, was answers that are “almost right, but not quite”. In 2026, 48 percent said they trust AI when they can easily check its answers.
Google’s DORA research found the same tension. In 2025, 90 percent of nearly 5'000 technology professionals used AI at work and more than 80 percent believed it made them more productive. AI use was now linked to higher delivery throughput, but still to lower stability. DORA’s conclusion is that AI works “as an amplifier, magnifying an organization’s existing strengths and weaknesses”.
Faster, or does it only feel faster
The best known study points the other way. In early 2025 METR, a research group, randomised 246 real tasks from 16 experienced open source developers working on large, mature projects, mostly with Cursor and Claude 3.5 and 3.7 Sonnet. The developers expected AI to make them 24 percent faster, and afterwards believed it had made them 20 percent faster. Measured, they were 19 percent slower.
METR has since said those results no longer reflect current tools. In February 2026 it published new data from 57 developers and more than 800 tasks. For developers from the first study it now estimates a speedup of 18 percent, and for new recruits 4 percent, with confidence intervals that still include no gain at all. The study also ran into the change it measures: 30 to 50 percent of developers said they chose not to submit some tasks because they did not want to do them without AI. METR calls its own figures “only very weak evidence for the size of this increase” and is redesigning the experiment.
The honest reading is that agents probably help most developers more today than in early 2025. Nobody has yet measured by how much with confidence, and how fast it feels is a poor guide.
Fewer doors for juniors
The pressure shows first at the entrance. Erik Brynjolfsson, Bharat Chandar and Ruyu Chen at Stanford follow payroll data from ADP covering millions of US workers. In their update of August 2026, with data to June, employment of 22 to 25 year olds in occupations exposed to AI, software development among them, stood 19 percent below where it would be had it kept pace with less exposed peers, while experienced workers showed no such gap. The difference came mainly from fewer hires rather than more departures, and it was concentrated where AI is used to automate work rather than to assist it. The authors call these early, descriptive signals, not estimates of cause.
Job postings tell a longer story. Indeed’s index of US software development postings stood at 78 on 2 October 2026, against 100 in February 2020 and a peak of 234 in February 2022. Most of that fall came in 2022 and 2023, before coding agents were in wide use, and since a low in May 2025 postings have risen by more than a quarter. The US Bureau of Labor Statistics still projects employment of software developers, testers and quality analysts to grow 10 percent from 2025 to 2035, with about 106'100 openings a year, driven in part by software for AI.
The skills that matter now
If agents write more of the code, the value moves to everything around it: saying precisely what should be built, reading code critically, debugging, testing, security, and designing systems that will still make sense in five years. Knowing the business a program serves now matters more than knowing a framework by heart.
For those still learning there is a catch. In a randomised trial published in January, Anthropic asked 52 mostly junior engineers to learn a Python library they had not used before. Those with an AI assistant finished about two minutes faster, a difference that was not significant, and then scored 50 percent on a quiz against 67 percent for those who coded by hand, with the largest gap in debugging. Participants who asked the AI for explanations, or only asked conceptual questions, kept most of their learning. In the authors’ words: “How someone used AI influenced how much information they retained.”
Developers sense this too. In the 2026 Stack Overflow survey, concern about losing job skills was the top reason given for avoiding AI at work or school.
The outlook
Our view is that the developer’s job is becoming more like an editor’s and an architect’s than a typist’s. There will be more software, written faster, and more of it will need someone who can judge it. The risk sits at the bottom of the ladder. If companies stop hiring juniors because agents do junior work, they will run short of seniors in a few years. The apprenticeship has to change rather than end: juniors who review, test and explain what agents produce, with seniors who check their reasoning.
That is how we work in Werk, our app platform. AI writes most of the code on a framework we built and trust. People decide what to build, design how it feels and review every release.
