Skip to main content
Post

Nobody Cares That Your Agent Did It

Graphify's "Nobody Cares That Your Agent Did It" is the fourth essay in its Better Software series. It's about staying accountable for code that agents write faster than anyone can read. This is a concept piece, not a product release.

Its argument is that agents change who types, not who answers. When a customer's checkout breaks, "the agent wrote that part" doesn't matter. The essay cites Moffatt v. Air Canada, LLVM's contributor policy, and Anthropic's own study of agent work hours to make the point that a person or team still answers.

For the context problem, the authors say the "why" behind a change is split across support tools, CRMs, billing, and Slack threads. They suggest planning for that fragmentation with three habits:

  • Every change links back to its why, so a teammate can check why it's there.
  • Every kind of data has an owner who keeps it current and settles conflicts between systems.
  • The team decides ahead of time which systems to watch closely, based on risk, and who to ask when one breaks.

The essay then borrows habits from operations teams, starting with blameless postmortems. It argues that accountability should mean the team can say what was meant, who decided, and what happened, and then fix it. It shouldn't mean someone read every line. That's a relevant read if you use Graphify to trace code with an AI assistant.

Loading replies