Simon Willison says AI agents can't be accountable

Willison argues an AI agent can never be a Directly Responsible Individual, per his July 12 post. DRI is Apple's term for the one person accountable when something ships broken.

That distinction matters for anyone deciding how much autonomy to grant a production agent. The agent can do the work, but it can't be accountable for the work, so a human still has to hold the bag.

The post reads as a counterweight to the 'let the agent own it end-to-end' pitch that treats full autonomy as the goal in itself.

The DRI framing is specific: one named person, not a committee, not a process. Willison's point is that swapping an agent into that role breaks the model at the root. Accountability needs a party that can be held to it.

Willison's post doesn't prescribe a replacement structure. It just states the constraint: the DRI has to be someone who can be found and asked what happened.