Control debt: every AI agent needs an identity
Companies are deploying agents into existing workflows without redesigning how decisions are made. The gap between the power of the technology and the visibility over its decisions has a name.
The share of companies abandoning most of their AI initiatives before production rose from 17% to 42% in a year, according to S&P Global Market Intelligence. In the MIT NANDA sample, 95% of generative AI pilots created no measurable financial value.
Yet the conversation inside most companies remains the same: "What tasks can we automate?" That question explains many of these numbers. Automation focuses on execution. AI changes something far more fundamental.
AI analyses. AI recommends. AI coordinates. AI influences decisions. The real question is not which tasks can disappear. The real question is which decisions can be delegated, which decisions require human judgment, and which decisions should be made jointly by humans and AI.
Where organisations get stuck
They deploy agents into existing workflows without redesigning how decisions are made. They invest in models before defining accountability. They create AI capabilities before defining governance. The result is predictable: 39% of organisations report that poor oversight of AI agents creates security and operational vulnerabilities. Only 19% actively measure the errors made by AI agents, even though accuracy is considered the most important performance metric by employees.
The technology is becoming more powerful. The visibility over its decisions is becoming weaker. That gap is what I call control debt.
An identity for every AI participant
Every employee in a company has a name, a manager, a mandate, permissions and accountability. Every AI participant should have the same. An agent with an identity can be governed, measured, audited and improved. An agent without an identity becomes shadow AI. It makes recommendations, influences outcomes, duplicates itself across departments and accumulates risk that nobody owns.
The companies creating the most value from AI are not asking departments to list tasks. They are mapping decisions. They identify which decisions AI can make autonomously, which decisions remain human, which decisions require human and AI collaboration. Then they build governance once and deploy it across the organisation.
The future competitive advantage will not come from having more agents. It will come from designing an organisation where humans and AI operate under the same architecture of accountability. AI is becoming a company architecture discussion.