"Human in the loop" is not a governance model
Three things about AI in program delivery that nobody puts in the business case.
3 — Ideas
- Your review capacity didn't scale with your output.
AI assistance multiplies drafted artifacts three to five times. The number of senior people available to review them stays exactly the same. Nobody budgets for review capacity when they budget for tooling, so the gap gets absorbed silently by the people least able to refuse it.
What happens next is never announced. Reviews become skims, or they become signatures. Both still get labeled "human in the loop."
- The status report was never the point.
It's the first thing every program tries to automate because it's the most hated recurring task. But the value of a status report isn't the document — it's that producing it forces someone to go reconcile reality. Chase the amber dependency. Ask whether "on track" means the same thing it meant last week.
Automate the byproduct, lose the forcing function. Status looks better; program gets worse. Takes about two quarters before anyone connects the two.
- AI made deferring a decision free.
The socially acceptable form of not deciding has always been asking for more analysis. Nobody gets criticized for wanting more information.
You can now generate a comparative options matrix, three scenario models, and a risk-weighted recommendation faster than you used to schedule the meeting where you'd have decided. The analysis is genuinely better. The decision still hasn't been made.
If your options analyses improved this year and your decision latency didn't, that's not productivity. That's Decision Debt accruing in a nicer format.\n\nThe debt is optional. The interest is not.
2 — Questions
How many AI-drafted artifacts hit your senior reviewers in a week — and how long would an actual review take?
If the second number exceeds their available hours, your oversight is already notional. You just haven't measured it yet.
Where's your Judgment Line — and who, by name, owns the output?
Not "the PMO reviewed it." A person who would be asked about it by name. Every artifact type where you can't answer that has no governance, whatever the policy document says.
1 — Thing to do this week
Write your own escalations.
Language models are trained toward agreeableness. Ask one to draft a risk memo and you'll get a slightly softer version of what you meant. Each instance is fine. Run it for a quarter across every risk memo and amber-status narrative, and your organization's risk signal attenuates — the escalations still go out, they just land with less force.
You can't spot this in any single artifact. You spot it when leadership stops reacting.
Escalations are the one category where the blunt human sentence is doing load-bearing work.
Full piece: [https://www.beadecisiveleader.com/insights/human-in-the-loop-not-governance-model] The Judgment Line, the A.R.C. Protocol, and the Four Surrenders are covered in full in Decisive AI, Volume 5 of the Decisive Edge series.
