On September 29 at OpenAI DevDay, one of the quieter announcements was a Meetings plugin for ChatGPT. It generates meeting notes and action items. It is in beta on macOS for Pro and Business. Alongside it came Team Tasks for recurring task assignment, Pages for collaborative documents, and ChatGPT inside Slack and Microsoft Teams.
This is good. I want to say that plainly before I say anything else.
Meeting notes are now free, or close to it, almost everywhere a leadership team already works. Nobody on your team needs to be the scribe anymore. Nobody needs to type while everyone else talks. That is a real gift to every company that runs a weekly meeting.
But I have run a weekly leadership meeting for years, and I can tell you where it breaks. It never broke because the notes were bad.
Here is the central claim. A list of action items is not a commitment. The failure does not happen in Thursday's meeting. It happens between Thursday's meeting and next Thursday's.
A perfect transcript does not change that. A perfect summary does not change that. What changes it is what happens to each item in the seven days after the meeting ends.
I have watched commitments die four different ways. Each one has a fix. None of the fixes is better notes.
Failure one: nobody confirmed they owned it
An AI listens to the meeting and writes, "Follow up with the vendor about pricing." It even guesses a name next to it.
But nobody in the room said, "That is mine." The name was inferred. The person it was assigned to may not have heard it, may disagree, or may think someone else was going to do it.
An inferred owner is not an owner. The item looks accountable on the page and is accountable to no one in practice.
The fix: the AI proposes, and a person confirms. Every to-do needs a human who said yes to it, in the meeting or right after.
This is how OTP handles it. OTP can record the meeting and propose to-dos from what was said. It does not enroll anyone. People confirm the to-dos that are theirs, and change or drop the rest. A to-do becomes real the moment its owner accepts it, not the moment a model writes it down.
Failure two: it lives in a doc nobody reopens
The notes go into a document. The document goes into a folder. The folder is never opened again until someone asks, three weeks later, "Didn't we decide this?"
I have hard evidence of how common this is, from our own company. At Sneeze It we have an agent named Crystal who reads our delivery boards on Trello. Her headline measure is board adoption: how many people actually touched the board in the last seven days. The number is 2 out of 11. Our goal is 5.
Two people out of eleven. That board is full of real work, well organized, and technically our system of record. It is not a system. A system of record that nobody opens is a filing cabinet.
Meeting notes have the same problem, only worse, because they multiply. Every week there is a new document. None of them points to the last one.
The fix: commitments should live where the team already looks every week, not in a document the team has to remember to find.
In OTP, to-dos sit inside the weekly Delta Meeting itself, next to the scorecard, the priorities and the issues. You do not go looking for them. They are on the agenda.
Failure three: nothing brings it back next week
This is the big one.
An item gets written down on Thursday. It is due Tuesday. Tuesday comes and goes. Next Thursday, the meeting starts fresh, with new topics, and the old item is simply not mentioned. It did not fail. It evaporated.
Most tools are very good at capturing and very bad at returning. The notes from last week do not walk into this week's meeting on their own.
The fix: an open commitment should come back automatically until it is done or someone explicitly releases it. Silence should not be a way to close something.
OTP carries open to-dos into the next Delta Meeting and keeps carrying them until they are done. Recurring work can be set as a repeating to-do, assigned to a person or to a team. Team Tasks in ChatGPT points in a similar direction for recurring assignments, which tells me the whole industry is waking up to this. The question is whether the item shows up in front of the leadership team, every week, until someone says it is finished.
Failure four: the boss's own items quietly get a pass
This one is uncomfortable, so most systems avoid it.
In almost every leadership meeting I have seen, the CEO's commitments are the ones least likely to be chased. Nobody wants to be the person who says, "You said you would send that last week." So the founder's items slide, and everyone notices, and the standard for the whole team drops to match.
An AI that writes notes cannot fix this. The notes will list the CEO's action items just as neatly as everyone else's. The problem is not recording. It is that the social pressure to follow up runs in one direction only.
The fix: the rule has to be structural, not social. The CEO's commitments get chased as loudly as anyone's, by the system, so no human has to be brave about it.
That is how OTP works, and it is not optional. The founder's open to-dos show up the same way a new hire's do. If a team does not want that, OTP is honestly not the right tool for them.
What this means for your meeting
Take the Meetings plugin. Use it. Let it write the notes. That part of the job is done now, and it costs your team almost nothing.
Then ask a harder question about the other six days. Who confirmed they own each item? Where does it live? What brings it back next week? And does the boss get chased too?
At Sneeze It we run this same discipline on our AI agents. Tally, the agent that pushes numbers to our scorecard, is built to never fabricate a value when a source is broken. It reports the gap instead. We hold people to that same standard in the Delta Meeting: say what is done, say what is not, and let the open items come back until they close.
Notes were never the problem. Follow-through was. It still is.
Frequently asked questions
What did OpenAI announce for meetings at DevDay 2026?
A Meetings plugin for ChatGPT that generates meeting notes and action items, in beta on macOS for Pro and Business plans. OpenAI also announced Team Tasks for recurring assignments, Pages for collaborative documents, and ChatGPT inside Slack and Microsoft Teams.
If AI can write meeting notes, why do action items still get dropped?
Because a list of action items is not a commitment. Items die when nobody confirmed they owned them, when they live in a document nobody reopens, when nothing brings them back the following week, and when the leader's own items quietly get a pass.
How does OTP handle to-dos from a meeting?
OTP can record the meeting and propose to-dos, and people confirm the ones that are theirs. Confirmed to-dos are carried into the next weekly Delta Meeting until they are done, and the CEO's commitments are chased as loudly as anyone's.
Sources
- DevDay 2026 Recap, OpenAI
- DevDay 2026 announcements and developer resources, OpenAI Developer Community
- OpenAI makes 20+ announcements at DevDay, 9to5Mac
Ask your AI what is still open
OTP's MCP server lets any MCP client read your to-dos, priorities and scorecard.
In Claude Desktop or Cursor or any MCP client, add this block:
"otp": {
"command": "npx",
"args": ["-y", "@orgtp/mcp-server"]
}
Restart the client. Then ask: "Use OTP to list every to-do from our last three leadership meetings that is still open, grouped by owner, with mine first."
Start free at orgtp.com. Every seat is free. You pay only for the AI you use.
The series
- OpenAI DevDay 2026 in one page: the five announcements that change how your company runs
- OpenAI just gave AI agents job titles. Now someone has to draw the org chart.
- ChatGPT will take your meeting notes now. Notes were never the problem.
- Always-on agents make your weekly meeting more important, not less.
- Every AI now plugs into everything. Your operating plan should be one of the things it plugs into.
All five on one page: What OpenAI DevDay 2026 means for how you run your company.
Series: What OpenAI DevDay 2026 means for how you run your company. Part 3 of 5.