May 2026:
1,741 Commits — Our Biggest Month

May 2026 was a breakout month. We delivered 1,741 commits across 16 active projects — about 2.3× our March volume — adding an estimated 444,000+ lines of code across 3,847 files. A team of three, working with its AI agents, sustained an average of 56 commits per day, peaking at 174 in a single day.

The surge was driven by parallel investment in an internal multi-agent development platform and continued client delivery. New work touched CI/CD automation across preview, staging, and production, database schema migrations, automated code review and formatting, and cross-repo synchronization — reflected in a stack that grew to seven languages, now including PostgreSQL/PLpgSQL and Terraform/HCL.

May pushed 7.42 billion tokens through Claude — our heaviest month yet — and 7.11 billion of them were read from cache (96%). Most of that volume is 1M-token context read straight from cache rather than recomputed: large contexts written once, then re-read across sessions. Even as token volume jumped ~70% over April, the cache-read ratio held — which is exactly why the marginal cost of each unit of work stays roughly flat as we scale.

Our development process is fundamentally different from traditional software engineering. We don't work from detailed specifications—instead, we co-develop with AI, iterating rapidly through ideas and implementations.

What might traditionally be called "bug fixes" are really iterative refinements—the natural result of building software collaboratively with AI assistants where the first implementation is a starting point, not a final product.

This approach lets us move faster, experiment more freely, and deliver features that truly match user needs rather than predetermined specifications.

• →Internal multi-agent development platform — phased build-out of AI coding orchestration
• →CI/CD automation — preview, staging, and production pipelines with automated redeploys
• →Feedback and backlog automation — cross-repo tracking and umbrella pin propagation
• →Database schema migrations and deeper Supabase integration
• →Automated code review and consistent formatting across the codebase
• →Expanded documentation across platforms and runbooks
• →Continued client delivery — admin tooling, data sync, and dashboard work

• ✓Deployment automation and one-command redeploys
• ✓Code formatting and style consistency (Prettier) org-wide
• ✓Cross-repo synchronization and submodule pin automation
• ✓Schema migrations and data-layer hardening
• ✓UI refinements across admin and client surfaces

Files modified this month, grouped by area of the stack:

Activity spread across 16 active repositories out of 203 total in our organization. Ranked by commit volume. Projects are anonymized — the description says what kind of product it is, never whose:

May proved we can scale output without losing the plot. Heading into summer, our focus areas include:

• Maturing the multi-agent development platform toward production use
• Broadening automated review, testing, and quality gates
• Deeper third-party and client integrations
• Continued infrastructure-as-code and database investment
• Sustaining velocity while tightening cost efficiency