Theo Browne examines Boris Cherny’s claim that coding is solved and Matt Pocock’s objection that engineering still requires strategic judgment. A truncated update notice in the Claude Code desktop app illustrates the distinction: source code can look plausible while producing an obviously broken experience that nobody has tested. The disagreement partly reflects different definitions of coding itself.
Theo Browne describes software quality as a cycle of planning, implementation and verification. A bad specification can produce bad software even when its instructions are followed accurately, while a vague specification leaves outcomes unpredictable. Running the result, checking it against user needs and feeding problems back into the plan matter as much as generating the implementation.
Theo Browne argues that teams must make verification straightforward for agents and people. Isolated desktop test environments can be harder to provide than browser previews, and difficult local setup encourages changes to reach review before anyone has tried them. Clear bug reports, repeatable tests and evidence of the running result can help, although an agent may still misinterpret its own recorded demonstration.
Theo Browne distinguishes rapid tactical implementation from strategic decisions about architecture and long-term codebase health. Agents can investigate alternatives and advise on plans, but people still need to design the development process and judge its results. Better test infrastructure and feedback loops improve both human and agent effectiveness without establishing that software engineering or bug prevention is solved.
Watch on YouTube
