DORA’s 2025 research found that 90% of respondents use AI at work, and that AI adoption still has a negative relationship with software delivery stability. Its summary is the clearest sentence written on the subject: AI doesn’t fix a team; it amplifies what’s already there. The gain at the keyboard is real. The bottleneck has moved to review, testing and stability, and that is where the work is.
It is also why we measure before we recommend anything. In METR’s controlled study, experienced open-source developers were 19% slower with early-2025 AI tools while believing they had been 20% faster. How a team feels about its tools is not evidence of what they changed.
Step one: the Engineering Throughput Review
Two weeks, at a fixed price quoted after a first conversation.
What you get
A measured picture of your delivery flow, from 90 days of repository and CI history: time to production, pull-request size and review wait, CI duration and failures, flaky tests, deployment frequency and rework.
How the tools are set up and used. Which tools and plans you hold, the context they are given about your codebase, how work is specified to them, and where their output stalls.
The codebase factors that slow people and agents alike. Test speed and reliability, coverage where change concentrates, local setup time, and the hotspots.
Conversations and observed sessions with your developers, reported in aggregate and never attributed.
Ranked changes to the codebase, the tooling configuration and the workflow,
with a drafted AGENTS.md for your main repository, a short guide to specifying
work for agents, and a 30/60/90-day plan with the measures to check it against.
The recommendations are tool-neutral: the review covers whatever your team
runs.
If the review concludes you should not do the work, or should not do it with us, you pay nothing.
It reviews the workflow, never the people. Nothing in it rates, ranks or compares an individual developer, and we agree that in writing before the first conversation. A request to do so ends the engagement.
Step two: the changes, at a fixed price
Putting the changes in place, and a hands-on workshop for your team on its own codebase, are quoted at a fixed price from the review. The work is in the code and the pipeline, not in a slide deck: the context file committed, automated review on every pull request, size limits and templates, a faster and more reliable test suite, and permission boundaries for what agents can reach.
We re-measure the same flow metrics afterwards and report what moved. We do not promise a productivity number, because delivery depends on everything else your team is doing at the same time.
Why us for this
The same tools, used on a regulated codebase where a mistake had consequences: AI review of every pull request in CI for compliance, credit-bureau reporting and security violations, with zero major compliance incidents. And four open-source releases shipped in July 2026 with agentic workflows, on libraries with over 81 million downloads, whose diffs anyone can read.
The failure mode with these tools is not that they are too slow. It is that they are too fast. Teaching a team to slow them down at the right moments is most of the job.
Who this is for
The CEO, CTO or VP Engineering who bought the seats, with a date attached: a seat renewal, a board review of the AI rollout, a hiring plan that assumes the gain, a customer security questionnaire asking how AI is used in your development, or an agent that reached something it should not have.
Tell us which tools your team has and what you expected them to change.