AI that makes one person faster can just as easily create a bottleneck for everyone downstream of them. The leverage I’m after isn’t personal productivity — it’s what changes when a whole team’s thinking is amplified together.
Treat it like onboarding, not like a tool
Bounded tasks first. Judgment observed. Scope expanded as trust is earned — not handed the keys on day one.
That’s the same sequence I’d use with a new teammate, and for the same reason: you don’t yet know where its judgment fails, and finding out on something that matters is expensive.
Context has to be explicit and shared
The failure mode isn’t the model. It’s context trapped in one person’s head or one person’s prompt history, which makes the output unreviewable by anyone else.
So: business logic gets defined before AI is asked to help build it, and the rules and skills that shape AI behavior are versioned alongside the code — not bolted on after, not living in a chat window.
When it’s an organization adopting, not one team
The premise that makes this tractable: AI amplifies the system it enters. It magnifies clarity or confusion, sound process or broken process, good habits or bad ones, with no preference between them.
That reframes the first question. It isn’t which tool — it’s what’s actually there to be amplified. Nothing about introducing AI improves a process that was already unclear; it just makes the unclarity move faster and reach further.
So the shape is a loop, not a rollout:
- Care — understand how the human and business system actually works. Where decisions get made, where ambiguity sits, where trust is thin, how work really flows as opposed to how the diagram says it does.
- Align — define what better means, for whom, and how you’d recognize it if it happened. Vague intent is the thing AI is worst at rescuing.
- Augment — introduce AI selectively, with the human decision and validation boundaries drawn before it goes in rather than after something goes wrong.
- Deliver — validate against real work and real friction. A demo proves the tool runs; it doesn’t prove the process improved.
- Evolve — build internal capability, so the organization ends up more able to think with AI rather than more dependent on whoever installed it.
The tool belongs at the bottom
Underneath all of it there’s an ordering I hold to:
Principle → Method → Playbook → Tool
Technology sits at the bottom deliberately. Models, vendors and platforms change on a cycle measured in months; the reasoning underneath them shouldn’t have to be rebuilt every time one does. A method that names a vendor in its second step has a shelf life, and it’s shorter than the problem it was meant to solve.
This is also why Care to Deliver isn’t an AI method. It’s the layer the AI method is derived from — which is what lets the derivative change without the foundation moving.
What it looked like in production
A multi-agent development workflow across Claude, Codex and Gemini with explicit roles — a context-extraction agent, parallel Challenger and Risk agents, a red-green-refactor cycle gated by human approval.
It cut time to market from months to weeks on the work it touched (full case study →). The part that mattered more: it changed how the broader team worked with AI, not just how I did.