So many groups are about to find that their software program practices are worse than they thought. Undocumented scripts, magical native setup, flaky exams, tribal-knowledge structure, obscure tickets, inconsistent naming, and “each senior engineer does it somewhat otherwise.” People simply realized to soak up it. Brokers expose this silliness instantly. An underspecified setting doesn’t create creativity; it creates rubbish. In the event you drop an agent right into a messy codebase and it flails, that’s not essentially an indictment of the agent. Usually it’s a really environment friendly audit of your engineering self-discipline. The repo is lastly telling the reality about itself.
Which is why I’d now say that my suggestion that AI coding requires builders to turn out to be higher managers was true, if incomplete. Sure, builders must turn out to be higher managers of machines. However extra importantly, they should turn out to be higher engineers within the old school sense: higher at specs, boundaries, “golden paths,” and so on. The agent period rewards self-discipline way over cleverness, and that’s most likely overdue.
So no, the large story of coding brokers isn’t that they’ll write code. Plain chatbots may already faux that half. The large story is that they’re altering what competent software program engineering seems to be like. Brokers reward precisely the issues builders have lengthy claimed to worth however usually prevented in apply: explicitness, consistency, testability, and proof. Within the age of brokers, boring software program engineering doesn’t simply scale higher, it does most every thing—collaboration, debugging, and so on.—higher.
