No memo, no decision is an agent control pattern

TL;DR: In a three-minute segment of this interview, Sharran Srivatsaa describes Acquisition.com's rule: no memo, no decision. Borrow it for consequential agent runs. Write the decision down before execution, when a reviewer can still change it.

Who should borrow this

This is useful for platform teams giving agents permission to change production state, spend money, delete records or communicate outside the company. It is less useful for a reversible local coding loop. The memo earns its cost when the action has a blast radius and a later reviewer needs to reconstruct why it was allowed.

Three minutes worth watching

• Around 28:04, Srivatsaa introduces a memo as a written record of the team's thinking.
• Around 28:39, he gives the operating rule: no memo, no decision.
• Around 29:07, he says the memo arrives before an executive meeting, so the meeting can work on questions rather than reconstruct context.
• Around 29:36, he treats the last 20 memos as institutional memory for a new employee.

My agent version is deliberately narrower. Before a run can delete data, change production state, spend money or communicate externally, require one short record: requested change, evidence, alternatives rejected, likely blast radius, rollback and accountable human. The approval attaches to that artifact. After the run, append what actually happened.

Make the gate mechanically boring. Store those six fields beside the run request, reject an empty evidence or rollback field, and bind the approval to that exact version of the record. If the request changes after approval, the gate closes again. That is a proposed control pattern, not a result demonstrated in the interview.

This is not a request for hidden chain-of-thought, and it should not become an essay the agent writes to exhaust the reviewer. The record exists to expose claims and choices that can be challenged. If its evidence is missing, fluent prose does not rescue it.

There is a useful tension a minute later: Srivatsaa suggests using voice chat with AI to reorganize rough thoughts into the memo framework. That can reduce drafting friction, but it can also turn uncertainty into cleaner-looking certainty. Keep source links and rejected options visible after the rewrite.

What the source can support

This is a self-reported operating practice in an 83-minute entrepreneur interview, not causal evidence that memos improve meeting quality or agent safety. I checked the English auto-generated captions and played the relevant frames around 28:14 and 29:35; I did not watch the full interview or test the memo method. Caption errors remain possible. The written guidance above is my proposed application, not a result claimed by the video.

Put the decision on paper before the agent acts, then make the outcome answer to the record.