A small group of London-based craftspeople aiming to become top 1% at using agents to build and maintain real, working products.
The goal
Get genuinely good at using agents to build and maintain real software — on mature, commercial codebases, at scale, without making mistakes.
That means going past prompt tips: running agents against real systems and keeping quality high as you hand over more of the work.
Who it's for
- 01
Engineers who struggle to get AI coding to scale beyond demos and prototypes.
- 02
Engineers building a software factory who struggle to hold quality while pushing autonomy.
- 03
Engineers already coding hands-off with parallel agents who want to push further.
What we study
- 01
Context layer
Curating exactly what the agent sees — the right context at the right time.
- 02
Planning
Build plans that don't fail, pushing the human higher up the abstraction layers.
- 03
Execution
Make it fast, efficient, reliable and parallel with remote environments.
- 04
Validation
Static analysis, conformance testing, model-based testing and formal methods.
- 05
Self-improvement
End-to-end feedback loops with automatic experiment selection and back-testing.
- 06
Evals
Empirically driven improvements instead of blind, gut-checked changes.
- 07
Tool agnostic
Use the best tool for the job. Learn principles rather than API surfaces.
What you get
A tight community
Small and curated. Everyone here is operating at the frontier — no spectators, no noise.
The techniques library
Free access to a growing library of techniques, patterns and ideas for autonomous, high-quality engineering.
Format
Study groups hosted in London, every other month. Hands-on, high-bandwidth, no fluff.
A high-signal WhatsApp group where the work-in-progress and the real findings get shared.
Come build with us.
We keep it small and high-signal. Tell us what you're building and how far you've pushed autonomy — we'll take it from there.