The Level 10 Meeting® has a tighter design than most people give it credit for. Segue. Scorecard. Rock review. Customer and Employee Headlines. To-Do list. IDS. Conclude. Ninety minutes. Same day, same time, every week.
Adding ChatGPT or Claude to that as an assistant is the most popular AI move in EOS® companies right now. It is also the move that most reliably degrades the meeting.
Here is what happens. The Integrator opens ChatGPT in a side window. The meeting starts. Someone hits a sticky point. The Integrator types "summarize the last three weeks of our retention data" while the meeting waits. The model returns a paragraph. The team reads it. Discussion continues. Multiply by four or five over the hour.
The meeting did not get faster. The meeting got broken.
What broke
Three structural things break when AI shows up as an in-meeting assistant.
The cadence breaks. L10® is built on cadence. Every section has a time box. The Segue is five minutes. The Scorecard review is five minutes. Rocks is five minutes. Headlines is five minutes. To-Dos is five minutes. IDS is the remaining sixty. The whole point is that the team gets through the first five sections fast enough that IDS gets the room. Live AI lookups inflate every section by 60 to 120 seconds. IDS shrinks. Issues do not get solved. Same issues come back next week.
The data trust breaks. The Scorecard exists so the team starts with shared truth. When the Integrator pulls a number from ChatGPT mid-meeting, half the room does not know if that number is from the Scorecard, from the team's CRM, or from the model's interpretation of a prompt the room cannot see. Trust in the data degrades. The next time a number is questioned, the meeting stops.
The seat boundaries break. Every L10® section has an owner. The Integrator drives. Each Rock has an Accountability Partner. Each Scorecard line has an owner. When AI is being typed into live by one person, that person becomes the de facto researcher for the entire team, which collapses the structure of who owns what. The Integrator is supposed to facilitate, not type into a model.
This is what we mean when we say AI as an assistant is not the same as AI as an agent. The assistant pattern is to type prompts during the meeting. The agent pattern is to have done the work before the meeting starts.
What an integrated L10® actually looks like
The agent pattern moves all of the AI work upstream of the meeting.
The Scorecard agent fills in last week's numbers before the meeting starts. Every line. If a number is missing, the agent has already escalated to the owner and the Integrator knows which lines are dirty.
The Rock review agent reads each Rock against its milestones and flags Off Track vs On Track in the prep doc. The Accountability Partner walks in already knowing the agent has flagged it. The five-minute Rock review actually fits in five minutes.
The Headlines agent reads the customer touchpoints, the inbox, the team Slack, and surfaces a short Customer Headlines and Employee Headlines list. The team reviews and adds, the team does not generate from scratch.
The To-Do agent reads the prior meeting's To-Dos against the team's task system and reports the completion rate. Done, not done, in progress. No one types this live.
The IDS agent does not solve issues. That is the human team's job. The IDS agent compiles the Issues List from the prior week of agent flags, manager flags, and Headlines, ordered by impact. The team does Identify, Discuss, Solve on a clean list.
The Integrator runs the meeting. Same Segue. Same time boxes. Same close. The model never gets typed into during the hour. If a question comes up that the team cannot answer, it goes on the Issues List or the To-Do list, not into ChatGPT in the moment.
Why this is a discipline problem, not a tools problem
The L10® meeting is one of the most disciplined operational rituals in small business. EOS® Implementers spend years coaching teams to hold the time box. AI as an assistant is so attractive in the moment that disciplined teams will degrade their own ritual to use it.
This is the same trap as Slack in 2015, email in 2005, instant messenger in 1999, and the BlackBerry in 2008. Every one of those started as a productivity boost and ended as a meeting destroyer until teams put up rules.
The rule for AI in the L10®: no model gets typed into during the meeting. All AI work is upstream. All AI output is in the prep doc. If you need something from a model mid-meeting, it goes on the Issues List.
How to retrofit this in a week
If your team has been doing the assistant pattern and the L10® has been slipping, retrofit in four moves.
Monday. Pick one section, usually the Scorecard, and have one person commit to fully pre-staging it before the next L10® with whatever AI tool the team already uses, even if it is just ChatGPT in a browser. The output goes into the prep doc 24 hours before the meeting.
Tuesday L10®. Run the meeting with the pre-staged Scorecard and the explicit rule that no one opens ChatGPT during the hour. Just one section, just one rule. Time the sections.
Wednesday. Debrief. If the meeting felt faster and cleaner, expand to Rocks next week. If it did not, find the prep gap.
Repeat. Each week add a section to the pre-stage list. Within four weeks the entire L10® prep is upstream and the meeting itself is dense and disciplined.
You do not need a fancy agent for this in week one. You need a person and a discipline. The agent comes later, once the pattern is in place.
FAQ
Can we use Claude instead of ChatGPT for L10® prep? Yes. Anthropic's Claude is strong on long-form reasoning and structured outputs. ChatGPT from OpenAI is equally capable. Either works. Pick one.
What about live transcription in the L10®? Transcription tools like Fireflies are not the same as in-meeting AI prompting. Transcription captures the meeting, which is fine. The trap is typing into a model live.
Does this apply to Quarterly and Annual sessions too? Yes, even more so. Quarterly and Annual sessions are higher stakes and longer. All AI work belongs in the pre-read.
EOS®, Entrepreneurial Operating System®, V/TO™, Level 10 Meeting®, L10®, Rocks™, Scorecard, IDS, and Integrator are concepts and trademarks of EOS Worldwide, LLC. This article is an independent practitioner perspective and is not affiliated with or endorsed by EOS Worldwide.