The Comprehension Obligation is the clause MING Labs writes into an agent’s job description. The human operator must stay able to reproduce that agent’s primary outputs unaided, within 30 minutes, rehearsed each quarter. Fail the rehearsal, and the agent drops one autonomy level until the operator is current again.[S1]
It is live in the Chief of Staff agent’s job description today and rolling across the rest of the fleet. We say that plainly because the alternative is a claim we cannot show, and a rule you cannot show is not a rule.[S2]
It exists to stop the quiet failure of enterprise AI: judgment migrates to the agent until no human can defend, change, or rebuild the work. Most tools sell that drift as the goal, which is judgment elimination. A hybrid organisation sells the opposite. The Comprehension Obligation is judgment preservation written into the contract, so the firm keeps ownership of the outcome no matter which agent, model, or vendor holds it next quarter.
It is the enforcement mechanism for Layer A of the ABC Framework , the judgment and accountability that stay human. Capability can move to the agent. Comprehension cannot leave the operator.
In practice it has three moving parts:
| Requirement | Standard | Consequence |
|---|---|---|
| Reproduce | Operator rebuilds the agent’s three primary outputs by hand, unaided, within 30 minutes | Proves the work is still human-owned |
| Rehearse | Scheduled comprehension check, once a quarter, logged per operator | Keeps the operator current |
| Demote | A failed rehearsal | Agent drops one autonomy level until passed |
This is what separates an agent that is genuinely on the org chart from one the firm can no longer explain. We extracted the obligation from running our own agents, including the failure that taught us the difference, when we fired an AI agent after thirteen days .[S2]
For the operational version of this, how a company actually demonstrates that its people can still do the work its agents do, and why the EU AI Act creates that question without answering it, see how to prove your team can still do the work your agents do .
