{"id":5436,"date":"2026-07-20T20:19:05","date_gmt":"2026-07-20T17:19:05","guid":{"rendered":"https:\/\/lasoft.org\/blog\/?p=5436"},"modified":"2026-09-02T21:38:34","modified_gmt":"2026-09-02T18:38:34","slug":"ai-in-software-development-survey","status":"publish","type":"post","link":"https:\/\/lasoft.org\/blog\/ai-in-software-development-survey\/","title":{"rendered":"AI in Software Development: What 30 Engineers Actually Reported in 2026"},"content":{"rendered":"
\n
Most AI productivity numbers you can find are either vendor marketing or self-reported guesses. We wanted our own, so we surveyed all 30 people on our delivery team: every developer, QA engineer, product manager, designer, DevOps engineer, and business analyst about how they actually use AI on client work under real deadlines.<\/div>\n<\/div>\n

What came back is the clearest answer we\u2019ve ever had to a question many people are Googling right now: how do software developers actually use AI tools? People use AI in production, for client work, and to meet deadlines.<\/p>\n

All 30 of our team members use AI. Each of them still reviews the output before it reaches a product. The average productivity gain is about 30%.<\/strong><\/p>\n

Key Findings<\/h2>\n
\n
\u2013 All 30 engineers surveyed (100%) use AI tools in daily client work.<\/div>\n
\u2013 All 30 (100%) review AI-generated output before it is merged or shipped.<\/div>\n
\u2013 Average self-reported productivity gain across the team was approximately 30%.<\/div>\n
\u2013 Gains ranged from roughly 0% to 5x by department and up to 12x on individual tasks.<\/div>\n
\u2013 Writing unit tests was the single largest gain: 4x to 12x faster (about 1 hour down to 5\u201315 minutes).<\/div>\n
\u2013 A SQL performance problem under load that previously took 3\u20134 days was resolved in about 1 hour a 10x improvement.<\/div>\n
\u2013 Approximately 60% of the team pays for premium AI tiers, some out of their own pocket, because free tiers produced materially worse output, including arithmetic errors.<\/div>\n
\u2013 ChatGPT was the only tool used in all 7 departments.<\/div>\n
\u2013 Team members who integrated AI deeply into technical work (code, logs, SQL) reported around +30%; lighter users reported +10\u201320%.<\/div>\n
\u2013 A minority of the team reported that AI made them slower, because prompt rewriting and output verification cost more time than it saved.<\/div>\n<\/div>\n

How Do Software Developers Use AI Tools?<\/h2>\n

Software developers use AI tools mainly to accelerate well-defined, repetitive, or research-heavy work: writing tests, generating scripts and migrations, analyzing unfamiliar code, drafting documentation, and writing SQL, while keeping architecture, complex business logic, and final quality review in human hands. Across our team, AI tools produced an average productivity gain of about 30%, and 100% of developers reviewed AI output before merging it.<\/p>\n

The Numbers We Actually Measured<\/h2>\n

Vague claims about tenfold productivity are precisely what we\u2019re trying to avoid, so let\u2019s be specific about what the survey found.<\/p>\n\n\n\n\n\n\n\n\n\n\n\n
Metric<\/strong><\/th>\nValue<\/strong><\/th>\nNote<\/strong><\/th>\n<\/tr>\n<\/thead>\n
People surveyed<\/td>\n30<\/td>\nAll gave detailed feedback<\/td>\n<\/tr>\n
Departments covered<\/td>\n7<\/td>\nDevelopment, QA, PM\/PO, DevOps, Design, BA, CRM PM<\/td>\n<\/tr>\n
Average productivity gain<\/td>\n~30%<\/td>\n~0% to 5x by department; up to 12x on specific tasks<\/td>\n<\/tr>\n
Team members who verify AI output<\/td>\n100%<\/td>\nNobody ships AI-generated work without reviewing it<\/td>\n<\/tr>\n
Using paid tiers<\/td>\n~60%<\/td>\nSome pay out of pocket<\/td>\n<\/tr>\n
Most-used tool<\/td>\nChatGPT<\/td>\nPresent in all 7 departments<\/td>\n<\/tr>\n
Biggest single win<\/td>\nWriting tests<\/td>\n4x\u201312x faster<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Average Productivity Improvement by Task<\/h3>\n\n
\n
Unit Tests<\/dt>\n
1200<\/dd>\n
<\/abbr><\/dt>\n
1200<\/dd>\n
<\/abbr><\/dt>\n
1000<\/dd>\n
<\/abbr><\/dt>\n
500<\/dd>\n
<\/abbr><\/dt>\n
500<\/dd>\n
<\/abbr><\/dt>\n
400<\/dd>\n
<\/abbr><\/dt>\n
350<\/dd>\n
<\/abbr><\/dt>\n
250<\/dd>\n
<\/abbr><\/dt>\n
70<\/dd>\n
<\/abbr><\/dt>\n
30<\/dd>\n<\/dl>\n

AI Tools Our Team Uses<\/h2>\n

We\u2019re deliberately not loyal to a single tool. Different models really are better at different jobs, and pretending one wins everything is how you end up with mediocre output across the board. Here\u2019s the actual stack, straight from the survey.<\/p>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Tool<\/strong><\/th>\nDepartments<\/strong><\/th>\nWhat we use it for<\/strong><\/th>\n<\/tr>\n<\/thead>\n
ChatGPT<\/td>\nDevelopment, QA, PM\/PO, DevOps, Design, BA, CRM PM<\/td>\nGeneral assistant, broad tasks, working with prior context<\/td>\n<\/tr>\n
Claude \/ Claude Code<\/td>\nDevelopment, Design<\/td>\nCode, legacy analysis, MCP integrations, best at understanding existing code<\/td>\n<\/tr>\n
Cursor<\/td>\nDevelopment, DevOps<\/td>\nIDE agent, code review inside the editor<\/td>\n<\/tr>\n
Gemini<\/td>\nDevelopment, BA, Design, DevOps<\/td>\nResearch and reasoning over large data<\/td>\n<\/tr>\n
Codex (OpenAI)<\/td>\nDevelopment<\/td>\nIDE integration, commits<\/td>\n<\/tr>\n
Lovable<\/td>\nDesign, PO<\/td>\nClickable prototypes<\/td>\n<\/tr>\n
Figma Make \/ Figma AI<\/td>\nDesign<\/td>\nPrototypes, design-system work<\/td>\n<\/tr>\n
Claude Code + Figma MCP<\/td>\nDesign<\/td>\nBulk design-system changes (tokens, components)<\/td>\n<\/tr>\n
Jira AI \/ Atlassian Rovo<\/td>\nPM, BA, QA<\/td>\nWorkflows, sprint summaries<\/td>\n<\/tr>\n
OpenAI API + MCP<\/td>\nDevelopment<\/td>\nCustom pipelines, auto-commits, browser automation<\/td>\n<\/tr>\n
opencode CLI<\/td>\nDevOps<\/td>\nConfigs, scripts<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\t\t\n\t\t\t\n\t\t\t\n\n\t\t\t\n\t\t\t\n\n\t\t\t\n\t\t\t\n\t\t\t\n\n\t\t\t\n\t\t\t\n\t\t\t\n\n\t\t\t\n\t\t\t\n\n\t\t\t\n\t\t<\/picture>\n<\/p>\n

In practice, this is how we choose between AI tools: <\/p>\n