AI-Assisted Development Scales Through Verification, Not Generation
A familiar pattern is emerging in AI-assisted software development.
Read articleArticles
Articles
Notes on dependable AI-assisted development and production architecture.
The Unreliable Narrator Problem in AI-Assisted Development
AI-assisted development has a structural flaw that competent output tends to hide. The system rarely says, “I do not have enough guidance.” It proceeds: it fills gaps, decides...
Read articleStage Separation: Why Software Construction Needs a Supervisor Pattern
Most AI development tools share a default shape: one model acting as planner, builder, reviewer, verifier, and project memory, all inside a single conversational loop.
Read articleTask Contracts as the Unit of Work in AI-Assisted Development
The prompt is the most familiar artifact in AI-assisted development. It is what users type, tools log, tutorials teach, and operators often mistake for the work itself.
Read articleMulti-Backend Routing Is Insurance Against Model Risk
Most teams building with AI are more coupled to a single model vendor than they realize.
Read articleTiered Retry Is a Failure Economy
These systems fail often enough that failure handling cannot be a secondary concern. It is a primary execution path.
Read articleMinimal Blast Radius Is What Makes AI Automation Tolerable
The system will sometimes be confidently wrong.
Read articleCurated Task Execution Beats Dynamic Planning in Real Work
The dream is seductive: give the system a goal and let it figure out the rest.
Read articleAnti-Patterns in AI-Assisted Development
A methodology needs diagnostic instincts, not just recommended patterns.
Read articleState, Resumability, and Why Chat Memory Is the Wrong Primitive
Chat is useful for thinking. It is weak as execution memory.
Read article