Blog
Writing
Notes on portfolio architecture and bounded autonomy.
Your fourth product should be your cheapest
If each product starts from a library rather than a blank repository, marginal cost falls with portfolio size. Almost every organisation says it wants this, and almost none can state its current reuse ratio — which is the first reason it never happens.
Generating code is the easy half
Producing a working implementation of a described feature is close to solved. Producing one that the next twenty changes do not have to fight is an architecture problem, and no amount of generation speed substitutes for it.
A model reviewing its own output is not review
Self-assessment shares the failure modes of the thing being assessed. Deterministic checks — types, tests, contracts, schemas — are the only part of a validation story that does not depend on the reviewer being right.
Autonomy is defined by what it refuses
Any system can be given more permission. The interesting property is the set of actions no configuration permits, because that set is what a security reviewer can actually rely on.
Rewrites are a Blueprint problem
A rewrite is expensive because the description of the product lives only in the code being discarded. Keep the description separate and versioned, and changing technology stops being an act of archaeology.
Provenance has to be captured, not reconstructed
Nobody can reliably reconstruct why a line was written six months after it shipped. Recording the Blueprint version, requirement, agent, model and approver at the moment of the change costs almost nothing and answers the question forever.
Stop managing prompts. Start managing software strategy.
Connect your repositories. Let Foundariq understand what you have already built. Then tell it what your portfolio should build next.
Read-only to start · Your code stays in your GitHub · [email protected]