Engineering throughput: the measurement behind 6.7×.
This comparison uses GitHub merged-PR records from one engineering team, covering May 2025 through September 26, 2026. It follows four team-defined phases as coding agents, connected workflows, autonomy, and Lit Machine became part of the engineering process.
What the chart measures
Each bar is the mean number of merged pull requests per full development week. The comparison includes 35, 14, 10, and 8 full weeks respectively. Chart labels round the means to 19, 38, 65, and 127; the multiplier is calculated from the unrounded first and final means.
| Phase | Period | Weeks | Mean PRs/week |
|---|---|---|---|
| Coding agents | May 2025–Jan 2026 | 35 | 18.8 |
| MCP + automation | Jan 2026–Apr 2026 | 14 | 38.4 |
| Plugins + autonomy | Apr 2026–Jul 2026 | 10 | 65.0 |
| Lit Machine | Jul 2026–Sep 2026 | 8 | 127.0 |
Which weeks are included
Six weeks identified by the team as planning or strategy periods are excluded: the weeks beginning April 28 and December 29, 2025; and April 6, June 15, August 3, and August 10, 2026. The final partial week, September 21–26, is excluded from the comparable full-week means. The underlying calendar history remains unchanged.
Quality and interpretation
The team observed a substantial improvement in quality as output increased. Human review and QA remained in the workflow. This is the team's qualitative assessment; it is not a measured defect-rate improvement.
The 6.7× result describes a change in engineering output across these phases. PRs measure repository activity, rather than customer value or production releases. The analysis does not isolate the effect of Lit Factory from changes in the work, contributors, models, or engineering practices, and it is not a forecast for every team.