Stack Overflow (2025) — Developer sentiment and usage of AI tools and agents

Summary:

This report summarizes how developers say they use AI tools and agents: how often, for what tasks, what they trust, and which products they use most.

It’s from Stack Overflow (Stack Exchange). While it has a commercial stake in dev tools, the findings come from a large survey, aiming to capture reported workplace and workflow changes.

Method: self-reported survey results shown chart-by-chart. Most questions have ~26k–34k responses, with smaller subsets for follow-ups (e.g., agent users). Different charts use different bases, so results aren’t all from a single consistent sample.

Key findings: AI is mainstream. 84% say they use or plan to use AI tools, and about half of professional developers use them daily.

Sentiment and trust are weaker than before. Roughly 60% feel favorable (down from 70%+ previously). More respondents distrust AI output than trust it, and very few highly trust it. For complex tasks, many rate AI as mediocre or poor—or avoid it entirely.

Use skews toward lower-risk work: searching for answers, generating content/synthetic data, learning concepts, and documentation. Far fewer use AI for higher-responsibility tasks like deployment/monitoring, planning, or code review.

A common complaint: AI is “almost right,” creating extra debugging. Even if AI writes most code, respondents still want humans involved for trust, ethics, security, understanding, and tough troubleshooting.

Agents appear early-stage. Many don’t use agents or don’t plan to. Just over half report at least some productivity benefit, more for individual speed than team collaboration. Top benefits: time savings and productivity; top concerns: accuracy, security, privacy. Commonly cited tools include ChatGPT, GitHub Copilot, Ollama, LangChain, Redis, Grafana/Prometheus, and Sentry.