Our method is the Deploy Ladder. Four phases: Map, Build, Wire, Ship. Three rungs, agreed before any code: an internal preview, staging on main, and production behind feature flags we own jointly with you. Every rung is a real URL, and the first flag flipped in production is the First Flip.
The first two days of Map are discovery, inside the fixed price: the tickets, work orders or dispatch calls the agent will handle, the decision points, the data. Output: the agent architecture and the three rungs.
Discovery inside the priceOrchestration, memory, retrieval and tool calls go in first. Stateless before stateful, deterministic before probabilistic. Every pull request deploys to a preview URL.
Working system foundationDomain agents land on the spine one at a time. Each arrives with its fail mode, its rollback path and the eval that proves it works.
Integrated agents + evaluationDeployed to your infrastructure or ours, behind feature flags we own jointly with you. The handoff box goes with it. Code is yours.
Deployment + 30-day supportEvery pull request auto-deploys to its own preview URL. Work in progress is something you can click, not something you hear about on a call.
Every pull request has a preview URL, and you have the link.
Merged work lands on main and runs on staging: the whole system, end to end, one step before production.
Main runs on staging, and the Friday demo can run there.
Production runs behind feature flags owned jointly with you. A feature goes live when its flag flips on, and comes back out the same way.
The agent runs in production behind a flag you co-own, with its fail mode, its rollback path and its eval in place.
The First Flip is the moment the first feature flag flips in production: a real piece of your workflow, running on the live system, behind a flag you can switch off.
It usually lands within ten working days of kickoff. We say usually because it is not a guarantee, and we won't call it one. What we do promise is that you'll see it happen on a URL, not read about it in a status report.
From kickoff to the first flag flipped in production. Not a guarantee.
Async first. What shipped, what's next, what's blocked. You read it when your day starts.
One call a week for the decisions that need a conversation. No standups for the sake of standups.
The production or staging URL, plus a Loom under five minutes. You click through the real thing yourself.
Decided before the agent sees real traffic. Usually that means handing the case to a person on your team, with the context attached.
The way back to how things ran before. Behind a feature flag, that's a switch, not a redeploy.
The test set and the scoring that show the agent does its job. The results go in the eval report in your handoff box.
We publish eval scorecards from shipped systems only with client sign-off. None are published yet. The open book tracks it.
Code is yours. It lives in your repo, and IP transfers on final payment.
The deliverable is a running system. Every engagement ends in a deploy.
The 30-min scope call is free and scopes the work. The first two days of Map are discovery, inside the fixed price.
Fixed scope, fixed price, SOW signed before any work. Change requests get their own SOW.
Nothing that only works in a meeting. If it can't run on Monday morning, it isn't done.
The First Flip usually lands within ten working days. We tell you that, and we don't call it a guarantee.
The engineer who scopes your system is the one who ships it.
30-min scope call, free. By the end you'll know what we'd build, which rungs it climbs, what it costs and how long it takes.