Larridin Data / Q3 2026
More AI. More output. A different kind of team.
What changed in engineering—and what leaders are changing next.
/ DATA
AI Impact inof merged pull requests were
AI-assisted in September
monthly AI coding-spend growth in September
down from 65% in July
of surveyed leaders reorganized
around products or outcomes
Selected findings · Open to everyone
How work gets done
Link to insightThe share of merged pull requests with AI involvement rose from 35% in April to 62% in September. The shift is visible in shipped work.
Average across measured companies · April–September 2026
What it costs
Link to insightMonthly growth in billed AI coding spend slowed from 65% in July to 13% in August and 11% in September, averaged across companies. Spend is still increasing, but far more slowly.
Output per engineer rose 33% from Q2 to Q3. Separately, AI spend per unit of output was 89% above May by September, with most of that increase occurring by July.
Average growth across companies with billed spend in every month from May to September 2026. Includes Claude Code, Cursor, OpenAI Codex and GitHub Copilot.
The separate spend-per-output index uses one scoring version: May = 100, September = 189. One later adopter raises that average; some companies were below their May level by September.
How teams are organized
Link to insight40.8%
of surveyed engineering leaders reorganized around products or outcomes in the past year.
Others changed manager spans, seniority and role mix, or created new roles in verification, agent platforms and AI enablement.
LDX3 New York · September 2026 · Approximately 98 respondents. Multiple selections allowed; percentages do not sum to 100%.
The complete picture
Get the analysis and metrics to guide decisions about AI investment, engineering performance and team structure.
AI Impact in Engineering Report
Q3 2026 · PDF · 21 pages
Local preview: emails are saved on this computer. The download is the final report.
Pull-request trends cover April–September 2026 at companies using Larridin Engineering Intelligence. Company averages give each company equal weight. Team changes come from a separate live, multi-select poll at LDX3 New York. Findings are descriptive and do not establish causation.
AI-assisted means a merged pull request carries any AI attribution, from direct or inferred evidence. Output measures shipped engineering work, not business value; it weights complexity and size and discounts missing tests and AI slop. Read the measurement methodology.
Output compares Q2 with Q3 on one scoring version, adjusted for the July 30 model change. Billed AI spend per output unit uses a fixed set of companies with billed spend in every month from May to September, indexed to May = 100. Coding-tool spend excludes ChatGPT and API spend; the week of September 28 is excluded because billing may be incomplete.
Company-based figures cover at least five companies and at least 25 engineers or users per company. Output per engineer includes engineers with at least 10 merged pull requests in the quarter. Larridin’s own engineering team is excluded. The poll had approximately 98 respondents, allowed multiple answers, and describes the leaders in that room—not all engineering teams.