Never Walk Into a Meeting Cold: Pre-Meeting Briefs From Shared Memory

You have a call in ten minutes. You open the calendar invite, recognize one name out of three, and start digging. The CRM has a contact record nobody has touched since March. The last call transcript is somewhere in your meeting recorder. There was a thread about pricing, but it happened in a chat with a colleague’s AI tool, not yours. The meeting starts. You wing it.

Everyone wings it. Not because the information is missing, but because it is scattered across five tools and nobody has twenty minutes to reassemble it before every meeting.

A shared memory changes the arithmetic. If the people, decisions and promises around an account have been saved to one engine as your AI tools worked, the brief is a single question to whichever AI tool you have open. Here is how that works, in enough detail to set it up today.

What a brief actually looks like

Here is the shape of one, for an external call. The names are invented. Your AI tool writes it, so the shape is whatever you ask for; this is what the prompt further down produces:

10:00 · Meridian Freight · quarterly review
Who's in the room
  Sarah Okafor, Ops Director. Your champion. Last contact 6 days ago.
  Tomas Rivera, CFO. First time on a call with us.
Where things stand
  Renewal in negotiation. Positive since the March pricing decision.
What we owe them
  Revised proposal (promised on the Mar 12 call). Due 4 days ago.
What they owe us
  Usage data for the Q2 review (Sarah, agreed Mar 12).
Since you last met
  They reported a problem with API limits (marked resolved).
  We decided internally to include the onboarding package.
What the engine doesn't know
  No record of a security review. If Tomas raises one, it's new.

Read that top to bottom in forty seconds and you walk in three steps ahead: you open with the overdue proposal before they chase it, you know the CFO is new to the relationship, and the API problem does not ambush you.

Notice the last section. Every line above it should come from a specific memory: a meeting, a person, a task, a decision. When your AI tool searches the engine, each result names the engine it came from, so the tool can say where a line came from and you can open it. Anything it cannot find goes under “what the engine doesn’t know” instead of being guessed. A brief you cannot trust is worse than no brief, and the gaps section is what makes the rest believable. It is also useful in itself: recurring gaps tell you which parts of the relationship nobody is saving.

Three meetings, three briefs

Ask for a different shape depending on the meeting.

External and sales calls get the account brief above: roles (champion, blocker, budget holder), where things stand, the last few interactions one line each, open commitments in both directions, live risks, and a suggested focus taken strictly from the open items. Staleness is information, so ask for it: “no contact in five weeks” belongs in the brief.

Manager 1:1s flip to the relationship, and they work best when the brief leads with your side of the ledger:

• What you owe them. The feedback you promised, the decision they are waiting on, the blocker you said you would clear. First, deliberately.
• Since your last 1:1. What they shipped; wins worth saying out loud.
• What they are carrying. Their open tasks, as context for support, not as a scorecard.
• Open threads they raised. The questions from three weeks ago that never got answers.
• Growth. Longer-arc career threads if the engine holds any, with a flag if it has gone quiet.

Keep the 1:1 brief to work artifacts, and frame it around what you owe and what to attend to. Say so in the prompt.

Team meetings get the shared version: who is in the room and why, open tasks and pending decisions across the group, what changed since this group last met, and a suggested agenda built from the open items.

Where the information comes from

The brief is not a feature you switch on. It is your AI tool reading a memory that your tools have been feeding all along. Context Engine remembers; your AI tool does the thinking.

What the engine holds, if you have been saving to it: people and companies, meetings with their attendees, decisions with the reason they were made, and tasks that are open or done, with an assignee and a due date. These link to each other, which is what makes “what do we owe Meridian?” a precise question rather than a vibe.

When you ask for a brief, a good AI tool does roughly this:

1. Finds out who is coming. If you have connected your calendar to the AI tool itself (Claude and ChatGPT both offer calendar connectors of their own), it reads the event. If not, paste the invite or name the attendees.
2. Searches the engine for each person and their company, by meaning, so “Sarah from Meridian” finds Sarah Okafor.
3. Pulls open tasks linked to those people, in both directions, and checks due dates.
4. Pulls decisions and the last meeting with them, and what was saved since.
5. Writes the brief, putting anything it could not find under the gaps heading.

There is also a head start you do not have to ask for. When an AI tool starts a session on a connection, it gets a brief from the engine: what each engine is for, what happened recently and what changed since it was last there. So it arrives at your question already knowing the week. If you want the plain-language version of this layer, start with What is a context layer?

What you tell your AI tool

A prompt that produces the brief above:

I have a call with Meridian Freight at 10:00. Using the Sales engine,
write me a pre-meeting brief: who's attending and their role, where
things stand, what we owe them and what they owe us (with due dates),
what changed since our last meeting, and a suggested focus. Only use
what you find in the engine or the calendar. Put anything you could
not find under "What the engine doesn't know". Keep it under 20 lines.

If you prep the same way every day, put it in the AI tool’s standing instructions (a Claude project’s instructions, or ChatGPT’s custom instructions): “When I say prep me for a meeting, write a brief in this shape from the Sales engine and today’s calendar.” Then “prep me for 10:00” is the whole request. Asking for “today’s meetings, one short brief each” in the morning gives you the digest version.

Closing the loop after the meeting

The brief is only as good as what was saved after the last meeting. So end the day with the other half:

That was the Meridian call. Save it to the Sales engine as a meeting
with Sarah and Tomas. Decided: we include onboarding in the renewal,
because it removes their main objection. I owe Tomas a security
overview by Friday. Sarah owes us the Q2 usage data. Save the decision
with its reason and the two follow-ups as tasks. Don't add anything
I didn't say.

Your AI tool saves a meeting, a decision and two tasks, each attributed to you and that tool, and tells you what it wrote. Tomorrow’s brief is better for it. If you have a transcript, drop it into Files or paste it in, and ask the tool to do the same from the transcript; see Meeting notes that go somewhere.

Tuning it

You tune it the way you would brief a new assistant: in the prompt.

Choice	Starting point	Alternatives
”Since last met” anchor	last meeting with these attendees	a fixed window
How far back	60 days	30 or 90
1:1 growth section	on, flags staleness	off
Length	compact, under 20 lines	longer dossier for board meetings
Which meetings	every meeting	external only, or 1:1s only

Failure modes, and what they mean

Attendees don’t resolve. The brief says it knows nothing about someone because the invite used a personal address or a name the engine has never seen. Save the person once, with their emails and company, and they are known from then on.

Briefs come back thin. Thin memory makes a short brief, and that is correct. A consistently thin brief for an account you care about is telling you that nothing about it is being saved. Fix capture, not the prompt.

The brief contradicts the CRM. Usually the brief is right and the CRM is stale. That is its own piece: the CRM that updates itself.

The tool fills gaps with guesses. Restate the rule in the prompt: only what it found, unknowns under the gaps heading. To check a line, ask the same question in Ask in the console; its answers name the memories they came from.

What changes

The obvious win is preparation time: twenty minutes of tab-hopping becomes a question and a forty-second read.

The less obvious win is what happens in the room. When you open with “last time you raised the integration timeline, here is where that stands,” the conversation starts three steps ahead. Your team notices too, because the follow-ups from the last meeting were saved, and so they get followed up. Managers may notice most: walking into a 1:1 already knowing what you owe your report changes the meeting from status collection to actual management.

Build this yourself

1. Create an engine with a purpose, for example: “Our sales relationships: accounts, the people in them, what we decided and what we promised. Not internal planning.”
2. Connect your AI tool with Can read and save. In Claude Code that is one command from the connect screen, such as claude mcp add --scope user --transport http ce-sales https://app.contextengine.com/mcp/c/<id>; for Claude and ChatGPT the screen shows which menus to open.
3. Connect your calendar to the AI tool itself, if it offers that, or plan to paste invites.
4. Save after every meeting with the closing prompt above, plus decisions and action items as they happen.
5. Ask for the brief before your next three meetings and adjust the prompt.

Give it two weeks before judging. The first briefs will have gaps because the engine is young, and the gaps section will tell you exactly what to start saving. See plans, or start free.