Product Agility
Building an AI-native platform

Small team.
Agent-scale ambition.

I’m exploring how AI agents can operate as a real delivery organisation — understanding requests, proposing improvements and building the platform autonomously inside a controlled environment.

↓
01 Why Product Agility

Time to learn.
Space to build.

Semi-retirement has given me more freedom to follow the things I want to learn, without losing the pleasure of doing useful work.

Renovating my house has introduced me to carpentry, insulation, roofing and the satisfaction of making progress with my own hands. It has also reminded me that every worthwhile problem rewards curiosity, patience and a willingness to rethink the plan.

A knee replacement added a rather more literal pause. Time away from the usual pace gave me room to recover, think and decide what I wanted the next chapter to contain. The answer turned out to be equal parts learning, making and seeing what AI agents can do when they are given a proper job — and sensible boundaries.

Product Agility has long been a home for my experience in strategy and product management. Now it is also my AI learning laboratory.

The long story, shortened ↗
01Make

Learning by doing, whether in a codebase or on a roof.

02Learn

Staying curious enough to be a beginner again.

03Orchestrate

Giving specialist agents the context, tools and guardrails to deliver.

02 The work in progress

A platform that can help build itself.

I’m building a configurable platform where ideas, context, comparisons and workflows sit alongside an agent organisation capable of improving the product itself.

User requests can become structured enhancement work automatically. Agents investigate, plan, implement and test — while approvals, permissions and audit trails keep people firmly in control.

Visit the platform holding page ↗
03 AI & autonomous delivery

Not one assistant.
An organisation.

I’m designing a system of specialist agents that can collaborate continuously. Each has a defined remit, limited authority and a clear hand-off to the next role.

SignalUser request

Feedback and enhancement ideas enter a structured queue.

→
Agent organisation
01Understand
02Design
03Build
04Test
→
ControlHuman approval

Evidence, permissions and audit history govern release.

01

Automatic intake

User needs become classified, prioritised improvement candidates.

02

Autonomous execution

Specialist agents coordinate work from discovery through testing.

03

Controlled change

Approval gates, scoped access and audit trails keep people in control.

04 The technology

Learning by
building.

A modern TypeScript platform, containerised for repeatability and connected to leading AI models and coding agents.

OpenAI CodexClaudeTypeScriptReactNode.jsPostgreSQLDrizzle ORMDockerViteAgent workflows
05 Notes from the build

What I’ve been
working on.

Short dispatches from an evolving platform.

01
Today’s build note

Giving the platform a customer with a clipboard

A repeatable customer-experience run followed real journeys across architecture, ideation, comparisons, diagrams and AI — keeping the evidence and turning the awkward moments into traceable work.

02
From the workshop

Giving a complicated diagram some manners

The process canvas can now arrange container-aware layouts, reverse changes, align selections and lock carefully placed records without losing the meaning underneath.

03
From the workshop

Putting work in its place

Process diagrams now understand participant pools, nested lanes and the important boundary between sequence flow within a participant and messages that cross between them.

06 The complete journal

Twenty-four days.
From the beginning.

24documented
Build Days