Delegation Is Now an Economic Decision

Ethan puts hard numbers behind this with what he calls the Equation of Agentic Work.

Three variables matter.

• Human Baseline Time: how long the task takes a person to do themselves.
• Probability of Success: how likely the AI is to meet the bar on a given attempt.
• AI Process Time: how long it takes to request, wait for, and evaluate output.

If review takes longer than doing the work yourself, delegation fails.

The GDPval study makes this real. Human experts took about seven hours to complete complex tasks. AI produced drafts in minutes. But reviewing the output still took about one hour. With GPT-5.2 winning or tying humans around 72 percent of the time, delegation only worked when review was fast and decisions were clear.

This is where many AI programs break. Teams assume iteration is free. It is not. Review time is real. Accountability is real. Someone still signs off.

Good delegation reduces review time. Bad delegation multiplies it.

Management Documents Become the Control System

Here is the practical takeaway most leaders miss.

The tools you already use to manage people are the best way to manage AI.

PRDs.Design briefs.Shot lists.Scopes of work.Five-Paragraph Orders.

These are not bureaucracy. They are translation tools. They turn intent into action.

Used as AI prompts, they dramatically improve first-pass quality and reduce back-and-forth. They also do something more important. They prevent agents from succeeding in the wrong way.

Without clear constraints, AI will happily deliver outputs that look good and create risk. Clear documentation creates lanes. It tells the agent what it can do, what it cannot do, and how success is judged.

As AI gets more powerful, this does not become optional. It becomes mandatory.

AI Can Help With Delegation, But It Cannot Own It

One important clarification.

Delegation engineering does not mean leaders must manually write every document or review every draft. AI can help draft specs. AI can ask clarifying questions. AI can critique its own work.

But AI cannot own intent.AI cannot accept risk.AI cannot decide when “good enough” is good enough.

Those decisions still sit with leaders.

The Bottom Line

If you’re leading AI adoption:

• CEO: This is a scaling behavior problem, not a tooling one. Results depend on how clearly work is delegated, reviewed, and decided, not which model you buy.
• CHRO: This is organizational design, not training. The critical capability is how work is structured, ownership is defined, and judgment is exercised across the company.
• CIO: This reduces risk by decentralizing learning. Clear delegation frameworks let teams experiment safely without turning AI adoption into an enterprise-wide failure mode.