writing
Effective service design and platform delivery.
Real-world, pragmatic takes on delivery architecture and continuous delivery practices in the era of agentic development. Each essay stands alone; together they trace a loop - design the boundary, agree it as a contract, deliver it, prove it continuously, then delegate against it.
headline · opens a four-part series
AI amplifies the architecture it finds
An optimist's case for sequencing. Agents will take real work off engineers and some of it should go today - but delegating before the structure is there just books the cleanup for later, against people who were never asked. The median architecture, MicroContext, and the two questions that survive delegation: which capability, and how the function operates.
- 01 AI amplifies the architecture it finds
- 02 Agents propose, contracts permit
- 03 Is the organisation the next codebase?
- 04 Can your pipeline carry an agent?
Services built to be consumed
Where you draw service boundaries and responsibilities decides who pays for complexity. "Let the API handle it" only works when the API actually owns the domain - core engine, domain service, versioned SDK, end-to-end verification.
boundary domain service versioned SDK
Read →How contracts earn trust
Versioned, executable contracts that evolved from test specs into trusted architectural assets delivered as a hands-on workshop that turns integration into an architectural style teams can make their own.
executable contracts parallel delivery
Read →The platform is the product
Treat delivery as something you architect, not something you push harder on. The System Version, the platform as a product, governance made evidence-based.
System Version evidence paved road
Read →The approval gate is not the problem
Enterprise CI/CD where approvals are permanent and push-to-production is never coming. The gate isn't what disqualifies you from continuous delivery - the manual labour either side of it is.
- 01 The approval gate is not the problem
- 02 CI is a practice, not a product
- 03 The gate, done properly
Can your pipeline carry an agent?
The loop, run as a pre-flight. Versioned inputs, an executable definition of good, evidence by default, an owner - the properties that make a pipeline trustworthy are the ones that make an agent trustworthy. Seven questions, each mapped to the step that answers it, and what it costs when the answer is no.
- 01 AI amplifies the architecture it finds
- 02 Agents propose, contracts permit
- 03 Is the organisation the next codebase?
- 04 Can your pipeline carry an agent?
by discipline
The same essays, grouped the way the work actually arrives. Several sit under more than one heading, which is usually the point.
Architecture
Where boundaries go, and who pays for getting them wrong.
Engineering
The practices that make the architecture survive contact with a team.
Delivery
Getting it into production, provably, under real constraints.
tools
Notes on tools I have put to work on my own projects. Shorter than the essays, and dated, because the tools move.
outcome Predictable delivery - trust earned through working code, not mandates.