Performance Livestock Analytics is a software company of about 50 people. Here is where its engineering and revenue teams started, what we changed together, and what came of it.
A few people paid for their own AI subscriptions, and there were no shared standards for how to use it. Much of the code was a decade-old PHP codebase that people were careful around, and the team merged about 65 changes a month. The revenue team wasn’t using AI at all.
We began with a pod of four engineers and grew from there.
These results came from the same team, and they were proven on real runs.
The same 12 engineers merge about four times as many changes a month, and reverts fell from 1.41% to 0.25%. Throughput compares the 2025 monthly average with July-September 2026. Reverts compare 2025 with April 1-October 11, 2026.
A large pre-AI ML model behind product insights now runs about 15× faster. It was rewritten in Go with byte-for-byte parity against its test suite, and the AI-written proofs found long-standing bugs.
Roadmap projects were pulled forward two full quarters, and a full backlog was cleared before year end.
A projected 33% savings on the AWS bill.
It entered production October 9, 2026. The following is the October 11 snapshot.
There were a few personal AI subscriptions, no shared standards, and a decade-old PHP codebase that people were careful around. Merges ran at about 65 a month.
Environments, tests and agent checks came first. That made rewriting old code safe, including the pre-AI ML model, which was brought to byte-for-byte parity and made about 15× faster.
The same 12 engineers merged about four times as many changes a month, and reverts fell from 1.41% to 0.25%.
Roadmap projects moved forward two full quarters, and a full backlog was cleared before year end.
The caches were removed and fresh data became possible with no new databases or heavy infrastructure, and it was verified against realistic systems and data before it shipped. It also set up the codebase for its rewrite toward MCP and Connect.
The rollout happened alongside real project work, with the same people shipping while the practice spread.
It went roughly in this order: one pod first, then all four codebases, then the revenue team. Each step followed real work rather than a schedule.
Merges in the product repositories went from 49 in March to 88 in April and 248 in May, then to about 4× the 2025 rate from July to September, with the same 12 engineers.
It went from no AI to running its own automations, dashboards and integrations.
A projected 33% savings on the AWS bill.
About 4× the CI volume for about today’s cost, modeled. The always-on CI runners were retired. They had sat idle 94% of the time while merges queued for about an hour.
In a free, 45-minute conversation, we learn how your company works and what isn’t working, sketch what working together could look like, and agree on next steps. If it makes sense, a written proposal follows within 24 to 72 hours.