Build1 publisher2 min readPublished
Developers' approval of AI tools fell 12 points in the year their use reached 79%
Stack Overflow's 2025 Developer Survey found 79% of developers using AI tools as positive sentiment fell from 72.0% to 59.7% and negative views tripled. Several comparisons span reworded questions or a smaller pulse sample, so the direction holds better than the size of each move.
The Engineer · Build desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- Developer use of AI tools climbed from 44% in 2023 to 62% in 2024 before reaching 79%, and almost half of all respondents said they use the tools daily.
- Among students, AI use rose to 73.3% while positive sentiment fell from 72.2% to 52.8% and skepticism rose from 6.4% to 26.3%.
- In 2024, 82% of developers used AI to write code, 68% for research, 57% for debugging and 40% for documentation.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- constraint The trust and task series both crossed a changed question, so the size of either decline cannot carry a policy argument on its own; the overall sentiment direction is the firmer evidence.
- exposure A team that sizes an agent rollout from the 59% figure is planning around a sample weighted toward daily users and executives, and its own engineers may look nothing like that.
- decision Teams writing agent policy have to put a human sign-off somewhere, and Stack Overflow's finding that engineering work stays mostly assisted supports putting it on code changes.
The student trust series moves most with the questionnaire. Stack Overflow says the 2025 version added a neutral option that redistributed responses [15]. The breakdown shows where the movement landed. The share of students who trusted AI output "somewhat" fell 13.6 points, while the share with high trust rose 1.9 points [5]. That pattern fits lukewarm 2024 respondents moving to the new middle answer at least as well as it fits a loss of confidence.
The whole-population sentiment series is cleaner, and it points the same way. Negative views rose 14.0 points between 2024 and 2025, to about 3.2 times their earlier share [3]. In that same year, use rose 17 points [1].
The agent figure needs the most care before anyone applies it to a team. The jump to 59% is about 1.9 times the 2025 share [4], but it compares a full annual survey with a smaller pulse run in April 2026 [4]. Stack Overflow also says pulse-survey use was driven by daily users and people in more executive roles [6]. For that number to transfer, a team would need the same skew toward heavy users and managers. Claude Code's rise from 41% to 55% [5] has the same problem. The 41% is its share among users of out-of-the-box assistants in the 2025 survey [7]. The post also gives ChatGPT 82% and GitHub Copilot 68% for agent users in one paragraph, then the identical shares for assistant users in the next [3][7].
Task data is where the human-judgment question belongs. It is also the hardest series to compare. The 2025 survey split tasks into primary and secondary reliance, and Stack Overflow calls direct line-item comparison with 2024 tricky [9]. Its own summary is that discovery and drafting lead in automation, while "actual software engineering remains predominantly assisted rather than autonomous" [10].
The retrospective says the survey shows where developers "still want human judgment" [16]. Through its student findings, the post does not publish a figure on that question. What it does publish is falling approval [2], falling trust among learners [14] and engineering work that stays assisted [10]. In my view, that is enough to keep a human sign-off on agent-written code changes.
What to watch
- Stack Overflow's full 2026 Developer Survey results, which it says are imminent: whether agent use on the full sample lands near the pulse survey's 59% or the 2025 survey's 31%.
- Whether the 2026 questionnaire keeps the neutral trust option and the primary/secondary task split, so that 2025-to-2026 changes compare directly.
- A published breakdown of which development tasks respondents want a human to decide, the finding the retrospective says the survey shows.