Your personal agent can read every email. It cannot read a decision nobody wrote down.

This will happen at a lot of companies this winter.

On Tuesday, the leadership team meets. They argue about whether to raise prices for new customers in January. After forty minutes they decide not to, and to revisit in March. Nobody writes it down. Everyone in the room knows.

On Wednesday, the sales lead's personal agent reads three customer emails asking about 2027 pricing. It finds a Slack thread from last week in which the CFO argued for the increase. It drafts a reply telling customers prices go up in January, and asks the sales lead to approve.

The sales lead is busy, sees a well-written reply that sounds right and taps approve.

The agent did not malfunction. It read everything it could reach. The decision was the one thing it could not reach, because the decision only existed in the memory of six people.

Agents see the argument, not the decision

Personal agents read what is written. In most companies, the argument is written, and the decision is not.

Email threads, chat channels and shared documents are full of positions: proposals, objections, "what if we," "I think we should." Those are what an agent finds. The moment the team settles the question usually happens out loud, in a meeting, and the record of it is a feeling.

Anthropic says it plainly about Claude Duet: it "can still misjudge what's urgent," and it may misread a thread. That is honest, but the bigger problem is the record. A perfect model reading an incomplete record still reaches the wrong conclusion.

What a decision record needs

A decision an agent can use has four parts:

What was decided, in one sentence. "We are not raising prices for new customers in January."
Who owns it. One name.
When it gets revisited. "March leadership meeting."
What it replaces. "This overrides the January increase proposed on October 14."

The fourth is the one people skip, and it is the one agents need most. Without it, an agent holding both the old proposal and the new decision has to guess which wins.

Meetings are where the gap is

Most decisions that matter are made in a weekly meeting. That is where a leadership team looks at the numbers, works the open problems and makes commitments. It is also where the written record is thinnest.

This used to be survivable. The people who needed to know were in the room. Now every person in the room has an agent working for them after the meeting ends, and none of the agents were in the room.

So the meeting has to produce a record the agents can read: decisions, commitments with owners and dates, the problems solved and the ones still open. A transcript will not do, because it is the argument again, forty minutes of it. Agents need the outcome.

The rule we follow

At Sneeze It, a decision is not real until it is in the record with an owner and a date. If it is not written down, the agents act as if it never happened, because to them, it didn't.

It sounds strict. In practice, it changes very little about the meeting and a lot about the week after it. When a person's agent drafts something that contradicts a decision, the person can see the decision right there, and so can the agent.

Sources

Give your team's agents the plan

Personal agents read whatever they are connected to. OTP is where your company's chart, priorities, scorecard and commitments live, and any MCP client can connect to it. In Claude or ChatGPT, add a custom connector with this address:

https://orgtp.com/api/mcp

Then ask: "Using OTP, what are my priorities this quarter, what did I commit to in our last meeting, and what is due this week?"

Start free at orgtp.com. Every seat is free, for people and agents. You pay only for the AI you use.

The series

All ten on one page: Personal agents are coming to work

Series: Personal agents are coming to work. Part 4 of 10. Read this post as Markdown: /blog/personal-agents-cant-read-unwritten-decisions.md.