Case study · Sep 7 – Oct 4, 2026
Four weeks with the Enso Method.
Five client engagements and eleven in-house products, all moving at the same time, with AI coding agents doing the building under the Enso Method. Measured from the git history and the agents’ own session logs.
- 1.08M
- Lines written in four weeks.Source code, tests and documentation. Generated files and data are left out.
- 3,632
- Commits.About 900 a week, each counted once, with merges removed.
- 16
- Projects moving at once.Five client engagements and eleven in-house products, across 106 repositories.
- 1
- Engineer behind all of it.Doug Kerwin, directing AI coding agents with the Enso Method.
The results
What four weeks produced.
Output held steady across the month. The lightest week still came to about 200,000 lines and 713 commits, and tests made up a quarter of everything written.
Lines of code written per week
Source code, tests and documentation, with that week’s commits below. Generated files are left out (555,391 lines over the period).
- Code
- Tests
- Docs
- Commits
Show the numbers
| Week of | Code | Tests | Docs | Lines | Commits |
|---|---|---|---|---|---|
| Sep 7 | 197,867 | 74,463 | 64,512 | 336,842 | 968 |
| Sep 14 | 91,754 | 67,001 | 38,686 | 197,441 | 713 |
| Sep 21 | 145,876 | 50,410 | 42,854 | 239,140 | 992 |
| Sep 28 | 165,932 | 76,004 | 62,006 | 303,942 | 959 |
| Four weeks | 601,429 | 267,878 | 208,058 | 1,077,365 | 3,632 |
Several agents, working at once.
Because each phase carries its own definition of done, agents for different clients can run side by side. The engineer moves between them, reviewing and deciding, rather than typing.
- 3.6
- agents working at once, on average, whenever any were working
- 4 in 5
- working hours had three or more agents running together
- 16
- the most working at one moment, in the week of September 21
The approach
The method does the heavy lifting.
The volume comes from how the work is organized, not from longer hours. Every project followed the Enso Method, the open-source methodology Enso Dynamics uses to build software with AI coding agents.
1 · Decide
For every project, the thinking happens first.
2 · Build
Each plan is cut into phases. An agent builds each one, in parallel wherever phases don’t depend on each other.
3 · Prove
Every phase is tested against real systems.
Each phase is one agent session. Dependent phases run in order; independent ones, and other projects, run at the same time: as many as sixteen agents at once in these four weeks.
Because the method is written down, it isn’t tied to one person. Every engineer who joins the work builds the same way.
How it was measured
Counted conservatively.
What counts
Every commit Doug authored for a client or an in-house product, by commit time in US Eastern. A commit that appears on several branches, or again after a rebase, counts once.
What’s left out
Merge commits, which repeat work already counted. Data files, snapshots, test evidence and anything else generated. Work by other people or their agents.
Agents in parallel
Taken from the Claude Code and Codex session logs, not from git. A session counts only while it is working on a request, never while it waits for a reply.
What it doesn’t show
Lines and commits measure volume, not value. Whether the software does what the business asked is what the method’s Prove step is there to check. No client, product or system is named, and the work is shown only as one total.
The Enso Method
The method is open source.
Everything in this case study was built with the same agent skills anyone can install for Claude Code and Codex. Read how it works, or talk with us about bringing it to your team.