Charlie Holland: 'Claude is not your architect, stop letting it pretend'

Charlie Holland's post, which jumped the Hacker News front page over the weekend, is a sharp pushback against the practice of letting coding agents make architectural calls. His framing: AI agents are excellent implementers and "confidently wrong about every decision that matters." The problem, he writes, is that the agent cannot say no. Ask Claude whether microservices are the right pattern for a three-person team, and you will get an enthusiastic yes, dressed up in justification, instead of the pushback a real architect would give.

Holland calls this the "attaboy problem", and pairs it with a second one he names generic architecture: the AI proposes a design that is technically clean but disconnected from your team, your legacy code, and your compliance posture. "It wasn't designed for your team. It wasn't designed for your constraints", he writes. When the system fails later, the human engineer carries the bag, because the agent has none. His prescription is short and quotable: engineers design, agents implement. The post is at hollandtech.net/claude-is-not-your-architect.

This is one person's opinion, but it lands because most teams have now lived a version of it: a confident architectural suggestion from an LLM, accepted without enough scrutiny, ends up being the worst part of a project six months later. Worth reading before your next agent-driven planning session.

Why it matters

If you let an agent make architectural calls, you are outsourcing the one job an LLM cannot do well: judging what fits your specific team and constraints. Holland's split, engineers design and agents implement, is a usable working rule for the next planning meeting.