Note
Convergence over capability
The next framework, agent, or tool is only useful when it resolves a concrete blockage better than the mechanism already in hand.
AI tooling makes it unusually easy to expand the solution space. There is always another orchestration framework, memory layer, agent pattern, routing technique, or retrieval system that could be relevant.
The cost is not only implementation time. Every additional option creates another branch in the reasoning process.
A practical counterweight is to optimise for convergence. Start with a bounded vertical slice and ask what is actually preventing it from becoming reliable. Add machinery only when it resolves that specific uncertainty or failure mode.
This does not mean avoiding experimentation. It means giving experiments a job: produce evidence that changes a decision.
A small mechanism that closes a loop is usually more valuable than a more capable architecture that creates five new open questions.