Module 8 — The AI Agent: Governing It · Lesson 8.6
Exec Meetings and Boardrooms
Seating AI executives, giving each a remit, and turning the minutes into real tasks
~12 min
What you'll learn
- Create exec personas with genuinely distinct remits
- Set up a boardroom, seat execs, and pick a chair
- Run one on demand before arming a cadence
- Judge whether a meeting produced anything, and fix one that did not
This is the most ambitious feature in the product and the one most likely to be either genuinely useful or expensively theatrical, depending almost entirely on setup. The difference is whether each seat can actually see something the others cannot.
What actually happens
You create a boardroom, seat two or more exec personas, and pick one as chair — usually the CEO.
On its cadence, the meeting runs. Each exec reports on its own area, using read-only data from the systems it can see. They discuss and challenge each other. The chair closes with action items, and those action items become real tasks in your workspace.
The minutes stay on the Exec Meetings surface: every exec's update, the discussion, and what it decided.
It is off by default. Unlike most of Kavanah's AI features, each meeting spends AI usage on a repeating cadence — a single meeting can be dozens of model calls — so an admin has to enable it under Settings → AI Agent, and convening one is governed under the AI-spend action category where you can set a daily budget.
The setup that decides everything
Create one persona per executive FIRST, and this is where the feature is won or lost.
Give each a role title — 'Chief Financial Officer', 'Chief Revenue Officer' — and, critically, restrict each one's capabilities to its own area. The CFO reads Finance. The CRO reads Clients and deals. The COO reads work and operations.
An exec with no scope has nothing to report, and a meeting of five personas all reading the same data is five paraphrases of one summary. The discussion is only interesting when the participants know different things.
Give each seat a remit as well: a sentence saying what it is accountable for. 'Pipeline, win rate and net revenue retention' tells the CRO what to lead with and what to be worried about.
Run one before you arm it
There is a Run now control. Use it.
A single on-demand meeting costs one meeting's worth of credits and tells you whether the setup produces anything. Arming a weekly cadence without doing this is buying a year of something you have not read.
Read the output critically. The questions: did each exec say something the others could not have said? Did anyone disagree? Are the action items things you would actually do?
If the answer to the first is no, the scopes are too similar. If the answer to the second is no, the remits are too vague — nobody has been given anything to defend. If the action items are generic, the meeting had no real data to reason from and you are back to the first problem.
Cadence and cost
Daily or weekly, at a time you choose.
Weekly is right for most businesses. Daily makes sense only if enough changes in a day for the meeting to have new content — and for most teams it does not, which means you pay for five meetings that say what Monday's said.
Set a daily budget on the AI-spend category so a misconfiguration cannot run away. This is exactly what that category and that mode exist for.
And review after a month: did any meeting change a decision? That is the same test as the State of the Business report, and the same fix applies — adjust cadence before cancelling, because a weekly meeting on a monthly business will always sound repetitive.
Doing it from chat
Everything here can be done conversationally: ask the agent to set up a boardroom, seat the execs, read the last meeting, or convene one now.
That is often the faster path for the setup, and it is definitely the faster path for reading — 'what did the exec meeting decide this week?' beats opening the surface and scrolling minutes.
Convening one from chat is still governed as AI spend, so if your policy parks that category, the request will queue for approval rather than starting a meeting. That is the intended behaviour and it is a good reason to keep a budget rather than an outright allow.
Set up a boardroom that works
- 1
Create exec personas with DIFFERENT scopes
CFO reads finance, CRO reads clients and deals. Same-scope execs produce five paraphrases of one summary.
- 2
Settings → AI Agent, admin only. Off by default because each meeting spends credits on a cadence.
- 3
Create a boardroom and press Run now
One meeting's cost to find out whether the setup works. Do this before arming anything.
- 4
Set a daily budget on AI spend, then arm it weekly
The budget is what stops a misconfiguration running away. Weekly suits most businesses; daily rarely has new content.
What to watch
- Distinctness of updates
- Whether each exec said something the others could not have said.
- Healthy signal: Always. Overlapping updates mean overlapping scopes, and the fix is the personas rather than the prompt.
- Action-item follow-through
- Share of generated action items that get done.
- Healthy signal: High. Action items nobody does mean the meeting is producing plausible work rather than real work.
- Credits per meeting
- What one meeting costs, and what the cadence costs per month.
- Healthy signal: Known, budgeted and proportionate to whether any meeting has changed a decision.
Key takeaways
- ·Execs must have DIFFERENT scopes — the discussion is only interesting when participants know different things.
- ·Give each seat a remit; a vague remit produces an exec with nothing to defend.
- ·Off by default, governed as AI spend, and a single meeting can be dozens of model calls.
- ·Run one on demand before arming a cadence, and read it critically.
- ·Weekly suits most businesses; set a daily budget so a misconfiguration cannot run away.
Last in this module: what all of this costs, and the surfaces that tell you.