Latest essayAutonomy Is a Blank Check
Every degree of autonomy you grant an AI agent is a constraint you remove from it, and architecture, following Gregor Hohpe, is the discipline of satisfying constraints, not of removing them. This essay argues that autonomy works like delegated signing authority: an approval limit set before the transaction, one that grows with a demonstrated track record and lives in the limit itself, not in reviewing every move afterward. The architect job is not to maximize the agent autonomy but to calibrate that limit, because autonomy is granted cheap and clawed back expensive.Practice9 min read
