Directly Responsible Individuals (DRI)

Simon Willison recently set off a small but critical conversation with a blog post questioning whether AI agents should ever be designated as a project’s Directly Responsible Individual (DRI). His answer was an unequivocal no — a stance that grows more important as AI weaves deeper into how we work.

The term DRI, popularized by Apple and later adopted by companies like GitLab, names the person who is ultimately accountable for a project’s success or failure. It’s not just about ownership; it’s an anchor of accountability. When something goes wrong, that individual must step up, explain, and bear the consequences — professional, legal, or personal. This requires uniquely human qualities: the capacity to understand repercussions and carry their weight.

Why can’t AI shoulder this responsibility? Imagine an AI-powered workflow that automatically processes customer orders. One day, a hallucinated reading of an email causes a million-dollar shipment to go astray. Who is to blame? The AI? You can’t reprimand it, dock its pay, or drag it to court. The buck will ultimately stop with the humans who designed the pipeline, deployed the agent, or were supposed to oversee it. Machines have no agency; talk of “AI accountability” dissolves at the organizational level.

This reveals a deeper trend: the more capable AI becomes, the sharper our accountability boundaries must be. As agents autonomously write code, answer emails, and make decisions, it’s tempting to hand them full authority over a domain. But that violates a management iron law: authority must match accountability. When AI exercises power without being accountable, systemic risk accumulates. IBM’s 1979 training slide — “A computer can never be held accountable, therefore a computer must never make a management decision” — resonates louder today precisely because AI now makes decisions at scale.

What does this mean for organizations and individuals? Companies must design “accountability architectures” alongside their AI workflows. Every automated pipeline and every agent needs a clearly assigned human DRI — someone with both the technical understanding and the authority to hit the brakes when the AI errs. For individual users, the same rule applies: you can use AI to draft reports or crunch data, but it’s your name on the file, and it’s you who must explain the logic behind the numbers. AI is an augmentation tool, not a stand-in.

A counterintuitive twist: the smarter the AI, the more demanding our role becomes. Superficially, AI promises to lighten our load. In reality, it demands sharper critical thinking and heightened vigilance. Like advanced driver-assistance systems that “free” the driver yet require instant manual takeover during edge cases, AI’s fluent output forces us to scrutinize errors hidden in plain sight. The DRI concept reminds us that this mental load is a non-negotiable cost — because we remain the human at the end of the accountability chain.

Willison’s brief reflection acts like a pin to the bubble of techno-optimism. Amid the din of “AI will replace everything,” we must remember that organizations are human constructs, and responsibility can only rest on human shoulders. Next time you’re tempted to entrust a project to an AI, ask yourself: who is the DRI? If the answer isn’t a specific person’s name, you might not be ready to launch.