Five phases, written down, predictable. No black box. You see the PRs, the evals, and the dashboards in real time. We tell you what broke before you find out from your users.
01
3-5 days
The principal reads your code and talks to the people closest to the problem. Engineers, PMs, customer support, whoever has the receipts. We come out with a single page of what is real, what is assumed, and what is broken.
Who’s involved
Principal, plus a research agent pulling docs and code maps in the background.
What you get
A one-pager. Problem statement, current state, three options with rough sizing.
02
2-5 days
We pick an option together and write it down. Hard scope, soft scope, explicit non-goals. A draft eval suite so we agree on what success means before we write a line of code. If we cannot agree, this is where we walk away clean.
Who’s involved
Principal leads. Eval agent drafts the test cases. You approve.
What you get
Signed scope doc, draft eval set, target metrics, a fixed quote.
03
2-6 weeks
Principal plus agent fleet in your repo. Daily PRs against a feature branch. We post a 5-line update every morning: what landed, what is in review, what is blocked. You see the work happen, you do not get a Friday surprise.
Who’s involved
Principal owns architecture and review. Implementation agents draft. Test agent runs the eval suite on every PR. Review agent flags regressions.
What you get
A working build behind a feature flag. Green eval suite. Latency and cost numbers from real traffic where possible.
04
3-7 days
Staged rollout. Internal users first, then 1%, 10%, 50%, 100%. Observability dashboards wired, rollback path tested, on-call rotation set. We do not flip the switch and disappear.
Who’s involved
Principal on call. Ops agent monitors error rates and latency. You sign off at each ramp.
What you get
Feature in production. Dashboards in your tool of choice. A written go-live note for the team.
05
1-2 weeks
Last phase is for your team, not ours. Runbook walkthrough, recorded code tour, eval suite explanation, a list of known gaps and follow-ups. We stay on call for 30 days after the engagement ends, no extra invoice.
Who’s involved
Principal pairs with your engineers. Docs agent compiles the runbook.
What you get
Runbook in your repo. Recorded handoff video. A 30-day post-mortem after we leave.
Not a sales engineer who hands off to a junior. Not a partner who shows up for the kickoff and disappears. The person who scopes the work is the person who ships the work.
A principal engineer reads every inbound. We reply same day on weekdays, with an honest read of whether we’re the right team for the work.