03/Method
Data first. Expert second. IT third.
Two halves of one method: deciding where AI agents pay for themselves, and building with AI agents as the engineering team.
03-A/Deciding
Where agents pay for themselves
Gate 00Before anything gets built, five questions
- Q1What problem are we solving?
- Q2Does it exist?
- Q3Is it the one we thought?
- Q4Does it need solving?
- Q5What is the cheapest fix that works?
Prove the problem before you price the solution.
Before anything gets built, five questions. What problem are we solving? Does it exist? Is it the one we thought? Does it need solving? What is the cheapest fix that works? Sometimes the answer is a process change and a spreadsheet.
Every initiative gets a number, an owner and a date.
A KPI tree linked to the P&L, with owners set through a RACI matrix. Without it, an AI program is a press release.
Decisions are the new bottleneck.
With agents, going from zero to a working MVP takes weeks. The value arrives only when leaders change roles and processes around the new tool.
Agents amplify management, good or bad.
If a manager can’t set a task and accept a result from people, agents won’t fix that. A team that plays together beats a team of stars.
The bridge rule
With AI you can now build a bridge overnight. But if nobody closes the old crossing, traffic stays where it was. If the river can be forded, you don’t need a bridge at all.
And don’t build on rented piers: today’s model prices are subsidized by investors and can change, so design for switching providers without a rebuild.
03-B/Building
How I build with AI agents
AI agents are my engineering team. I design, decide and check the result. In Evolution Workbench, 103 automated test scenarios gated every step and 46 decisions were written down. The Rat Box was built by AI agents from a 202-question design interview. I set up the same way of working for clients: a compact team that plays together instead of a team of stars.
- 01Director Task or decision Often dictated, with a screenshot. The human owns the why.
- 02Crew Design interview Open questions come one at a time, with the agent’s proposed answer first. 200+ questions before code on a new product.
- 03Crew Build Agents write the code against the spec and the decision log.
- 04Gate Automated checks Test scenarios gate every step. A failing gate stops the line.
- 05Crew Report What was done, what was measured, and what nobody has checked by hand yet.
- 06Director Review and decide The human tries it, accepts or sends it back. The decision goes into the log.
Decision log: every “why” is written down, so the next loop starts from facts, not memory.
The kit
What the AI Engineering Team Setup leaves behind
Set up on your first real project, so the method survives after I leave.
AI Engineering Team Setup- Design-interview template200+ questions answered before any code
- Decision-record logevery “why”, with the date and the alternatives
- Test-gate checklistwhat must pass before the next step starts
- Agent roles and handoffswho builds, who checks, who reports
- Cost guardrailslimits on spend per request and per day