Claim
AI agents should never be assigned as the Directly Responsible Individual for a project. The DRI role requires someone who can be held accountable for outcomes, and machines cannot bear accountability the way humans can.
Mechanism
The DRI framework requires a named party who owns success or failure externally. Accountability involves more than decision-making: it includes the social, legal, and reputational capacity to answer for what was done. A machine that makes a decision cannot be fired, sued, or questioned after the fact. Assigning DRI to an agent creates a structural gap: decisions get made but no accountable party exists to own them. Willison traces this principle to a 1979 IBM training slide: "A computer can never be held accountable, therefore a computer must never make a management decision."
Conditions
Holds when: projects carry real stakes and require someone to own failure externally. Organizations using DRI frameworks, OKRs, or RACI models.
Fails when: the "project" is a fully automated pipeline (batch job, scheduled cron) where the human who designed and deployed the system is the implicit DRI. Automation is not the same as delegating accountability.
Evidence
Willison wrote:
"An agent should never be considered the DRI for a project — that's something that feels uniquely human to me, because humans can take accountability for their actions where machines cannot."
He cites the IBM slide as precedent:
"A computer can never be held accountable, therefore a computer must never make a management decision."
Signals
- Post-mortems name human owners, not systems or agents.
- Teams report agent outputs as recommendations, not decisions.
- When an agent's output causes harm, a human accepts responsibility.
Counter-evidence
In high-frequency automated systems (fraud detection, content moderation, algorithmic trading), machines make consequential decisions continuously with no per-decision human review. The accountability falls on the team that built and deployed the system, diffused across the organization. The DRI principle applies to project-level accountability; it does not resolve accountability diffusion in always-on automated systems.
Cross-references
- Related to Frontier models capable of multi-hour autonomous runs require commissioning, not steering: frontier models running 9-hour autonomous projects still require a human commissioning owner, not a machine DRI.
- Related to Unattended harness loops accumulate defensive complexity until codebases assume machine participation: Ronacher's concern about hands-off harnesses producing worse outcomes implies the same structural gap.