Thoughts on AI-native development, machine-readable architecture, and building systems that AI agents can safely operate.
Two papercuts AI agents kept hitting in real projects: sysmara db generate logged success but produced no files, and every type: string field collapsed to VARCHAR(255). Both are fixed in v0.8.3.
Read more
SysMARA v0.8 fixes the gap between generated code and working code. Status transitions, associations, pagination, actor extraction, and proper error codes — all generated from your YAML specs.
Read more
We gave Claude the same 5 backend tasks twice — once with a plain Express prompt, once with SysMARA specs. Then we counted the invariant violations. 50% vs 0%.
Read more
We built an e-commerce system with 2 modules, 3 invariants, 1 policy, and 1 flow — then traced exactly how SysMARA makes every constraint visible to AI agents.
Read more
Step-by-step tutorial: create a fully working CRUD API backed by MySQL using SysMARA. From empty directory to running server in under 5 minutes.
Read more
SysMARA v0.6.2 changes scaffold generation from TODO stubs to working ORM-based implementations. Capability handlers, policy enforcers, and invariant validators now come with real logic out of the box.
Read more
SysMARA v0.6.2 integrates database selection, Docker environment, and environment files directly into the init command. One command gives you a fully configured project with a running database.
Read more
Traditional ORMs hide SQL behind abstractions designed for humans. SysMARA ORM does the opposite — every query is scoped to a capability, every operation is logged as structured JSON, and migrations know which invariants they affect.
Read more
SysMARA v0.5.1 introduces the scaffold generator: YAML specs in, typed TypeScript implementation files out. No more empty app/ directories.
Read more
How a double path-join bug in the capability compiler caused generated files to recreate the entire filesystem tree inside app/generated/. A postmortem on path handling.
Read more
Convention over configuration breaks down when the consumer is not human. Decorators, middleware chains, and implicit routing create blind spots that AI agents cannot navigate.
Read more
Machine-readable does not mean "has an API" or "uses JSON." It means the system's architecture is a formal, queryable artifact that tools and agents can traverse.
Read more
Business rules scattered across middleware, services, and database constraints are invisible to AI agents. Making invariants named, described, and bound to entities changes everything.
Read more
AI agents will modify any file you let them. Safe edit zones create explicit file ownership boundaries that prevent accidental damage without blocking productive work.
Read more
Instead of starting with routes or controllers, start with capabilities: named operations with typed inputs/outputs, entity bindings, policy gates, and invariant checks.
Read more