Tales from the 2026 Developer Survey results

Erin Yeppis says 83% of developers use AI at work, with 73% of those users doing so daily, primarily through coding assistants and general-purpose chatbots. AI agents are used by 66% of respondents, compared with 63% using general AI chatbots, and roughly 60% use between two and five AI tools. The leading LLM companies remain the frontier vendors: Anthropic and OpenAI each reached 74% usage, with substantial overlap. Open-weight providers remained steady year over year rather than gaining or losing significant share.

When developers changed AI tools during the past year, only 6% cited cost, making it one of the lower-ranked reasons. Although 33% said they personally monitor usage because of workplace token costs, 75% change prompts for quality reasons, compared with 38% who change them because of cost. Trust is conditional: 48% trust AI output when they can verify it, while only 7% trust it for many tasks and decisions. Source attribution matters to 79% of respondents evaluating AI-generated answers.

The most costly AI-related disruption reported by companies was plausible but wrong code, repeating a leading problem from the previous survey. Nearly-right code was also the most common source of friction, followed by security concerns. Ryan notes that this undermines the hope that AI-generated code could be accepted without human validation. Erin connects the problem to documentation and context: developers need attributable, current information to help agents produce correct code. AI use for documenting code nevertheless increased, from 33% mostly or partially using it last year to 51% currently using it.

Seventy-five percent of developers spend at least five hours a week searching for information; 33% spend five to nine hours and 20% spend 10 to 14 hours. Erin compares this with 2023, when 63% reported spending 30 minutes to more than 120 minutes per day searching, suggesting a possible but modest improvement. Important context arrives after work has begun for 79% of users: 43% experience this sometimes and 12% very often. Context remains fragmented: 54% keep separate notes, 45% start fresh AI sessions, 35% manually find files, and 33% copy and paste between tools.

Among people visiting Stack Overflow, 78% read posts and discussions, while 31% ask questions and 34% answer them. Sixty-five percent say they can find answers without posting, often through search or an AI agent, and 45% use AI to get enough information to move forward without further verification. Another 36% lack time to read the surrounding context, while 21% avoid posting because they fear judgment or criticism. Only 16% say they visit Stack Overflow more often to verify AI output; last year, 35% said they had visited because they encountered an AI-related problem.

Developers continue learning to code or learning new technical skills, primarily through technical documentation. Fifty-three percent now use AI tools to learn to code, and 51% use them to learn coding for AI. Erin says AI has not replaced human direction: 44% decide what to learn next through coworkers and peers, and 33% through managers. The survey also identifies JavaScript as the most-used language for the twelfth consecutive year at 62%, followed by Python at 58%; Rust grew to 18% usage and remains one of the most desired languages.

Postgres remains the leading database, used by 58% of respondents, and the survey now includes time-series comparisons showing its long-term movement against MySQL since 2018. In the new code-testing category, ESLint led at 37%, followed by Postman/Newman and Pytest. For AI orchestration, LangChain led at 30%, followed by the OpenAI Agents SDK and Terminal. Erin added testing because nearly-right AI code and fluctuating trust make independent validation increasingly important; she hypothesizes developers may prefer separate testing tools rather than asking the same AI system to catch its own mistakes.