Thirty days is not a marketing number. It's a constraint we impose on ourselves because it forces the right behavior: pick one metric, re-engineer one workflow, and put something real into production fast enough that the feedback loop stays tight.
Week 1: Diagnose, don't discover
We don't open with a discovery phase that bills for months and concludes with a slide deck. We map the target workflow, its data, and its systems in days, and leave the week with a single prioritized outcome and the definition of done attached to it.
Weeks 2–3: Build against production
The agent is wired into the real stack from the start, behind guardrails, working on real data. Building against production from day one is what kills the demo-to-deployment gap before it can open.
Speed comes from scope discipline, not from cutting corners. We compress the timeline by narrowing the target, never by skipping the controls.
Week 4: Prove it and hand over the keys
By the end of the month there's a metric that moved, an audit trail that explains every decision, and an operator who trusts the thing enough to leave it running. Then we widen the scope, but only against results we can already point to.
- One metric, named on day one.
- Production data and guardrails from the first build.
- Security and auditability in scope, never bolted on after.
- A working outcome before we talk about phase two.
