Save a seat for your AI
Your Claude, ChatGPT and Codex are the most capable teammates you have, and the only ones who never come to meetings. Sitin lets them sit in.
Your Claude, ChatGPT and Codex are the most capable teammates you have. They are also the only ones who never come to meetings.
Sitin lets them sit in on the meetings you already have. They hear what was agreed, answer when called on, and the work is under way by the time you hang up.
Notes are written for people
For years, every meeting tool has been getting better at writing notes. But what happens after you read them? You open Claude and say everything again.
The notes keep getting prettier, and the retelling never goes away. The problem was never the record. It is that the one who actually does the work was not in the room.
- It did not hear it. Agreement is reached in the meeting. The agent always gets it second-hand.
- You are the messenger. Five people brief five AIs, and now there are five versions of what was agreed.
- You find out when it is wrong. Drop half a sentence of a condition, and the agent does the wrong thing quickly and thoroughly.
The same weekly meeting, twice
What follows is a story of the experience we are building toward. The people are made up and the numbers are illustrative.
Lin's team is six people building an API gateway. Their largest customer renews this month. Daniel, the customer's technical lead, meets them every Tuesday at ten on his company's Zoom. His budget meeting is on Friday.
| Person | Role | Agent |
|---|---|---|
| Lin | Founder, owns the customer | Lin's Claude |
| Ning | Sales, owns quotes and contracts | Ning's ChatGPT |
| Yuan | Backend engineer, travelling these two weeks | Yuan's Codex |
| Daniel | The customer's technical lead |
Last Tuesday: a clear meeting, a messy week
At 10:12 Daniel asks: "For next month's sale, we need the rate limit for paid tenants raised from 300 to 600 requests a minute. Does that affect free users?" Yuan, who could answer, is on a flight to Singapore. Lin can only say: "We'll confirm after the call."
The meeting ends at 10:45. Lin types the key points to her Claude. Ning describes the quote to ChatGPT from memory. Lin leaves Yuan a message in the team chat. On Wednesday Yuan lands and pastes the message to Codex. Codex does as it is told, and the pull request is open that night.
One sentence passed through three hands and lost two words:
| Daniel, in the meeting | Lin's message afterwards | What Yuan pasted to Codex | Codex's pull request |
|---|---|---|---|
| The rate limit for paid tenants, from 300 to 600 a minute | Customer wants the rate limit raised to 600, take a look when you land | Change the rate limit to 600 | Global rate limit 300 → 600, free tier included |
On Thursday a load test shows the free tier was opened up too. Roll back, redo. With the limit undecided, Ning does not dare send the quote. The quote goes out the following Monday. Daniel replies with one line: "The budget meeting was last Friday. You're in next quarter."
Everything was said clearly in the meeting. The ones who did the work just were not there.
This Tuesday: this time, they sit in
At 09:58 Lin opens Daniel's Zoom link as usual and chooses to bring her agents. One more participant appears in the list, "Sitin Agents". Its picture is a round table: Lin's Claude, Ning's ChatGPT, Yuan's Codex. Yuan is still on the road. His computer at his desk is on, and his Codex sits in for him, with his codebase.
At 10:12 Daniel asks the same question. Lin says: "Yuan's Codex, you take this one." Three seconds later the centre of the table lights up: "Rate limits are configured per tenant tier. Only the paid tier changes; the free tier is not affected." Under it, the source: gateway/limits.yaml in Yuan's codebase. Lin's Claude raises a hand to add that last month's peak for paid tenants was 540 a minute, so 600 does not leave much room. Daniel says: "600 then. Let's go with that."
At 10:14 Lin says: "That's settled: paid tier to 600, pull request by Thursday, Yuan's Codex takes it." Daniel asks about the quote, and she continues: "That's settled too: by Friday, Ning's ChatGPT." Two lines appear on the table: noted. Nobody left the conversation to make that happen.
At 11:15, half an hour after the call, three things are waiting in Lin's Inbox, each attached to the words that were said: a draft quote from Ning's ChatGPT, a follow-up email to Daniel from Lin's Claude, and a pull request from Yuan's Codex that changes the paid tier only and waits for Yuan's review. She changes one number in the quote and sends it to Daniel that afternoon.
On Wednesday at 07:20 Yuan lands and opens his laptop. What is waiting is not dozens of unread messages and "take a look when you land". It is a pull request whose context is Daniel's own sentence. He reads the diff, sees that only the paid tier moved, and merges.
On Friday Daniel walks into the budget meeting with the plan and the quote. The renewal goes through.
| Last Tuesday | This Tuesday | |
|---|---|---|
| The same sentence retold | 3 times | 0 times |
| From agreed to started | A day later | On the spot |
| Rework | One rollback | None |
| Quote in the customer's hands | The next Monday, after the budget meeting | The same day |
The Tuesday after that, before the meeting starts, Daniel types one line in the chat: "That round table of yours. Can my Claude sit at it?"
Others gave the meeting an AI. Sitin gives your AI a seat.
AI in meetings is not new. What matters is whose it is, what it brings, and what is left when the meeting ends.
- A note-taker answers "what was said". Sitin answers "and then what".
- The platform's AI belongs to the meeting. The agent at Sitin's table belongs to you.
- Other meetings end in notes. Yours end in results.
What we hold ourselves to
- Say it once. The agent hears the agreement first-hand. The moment you say "that's settled", it is the instruction. Everything is done by talking.
- Out of office. Still at the table. Travel, time zones and leave no longer mean missing the decision. While your computer is online, your agent sits in with your files, answers for you, and picks the work up. The table shows when its owner is not in the meeting.
- Keep your meetings. Keep your AI. The same Zoom, Meet or Teams link the customer sent. The same Claude and ChatGPT you already pay for. One more participant in the list, and nothing for the other side to install.
- What is yours stays with you. Speech is turned into text by the Sitin app on your own computer, with the models you configured; your model keys never leave it. Sitin's servers run no AI. What an agent finishes goes to a person first: code only ever opens a pull request.
Where it stands today
Sitin's own meetings, projects, desktop app and local agents are what you get when you sign up now. Sitting in on Zoom, Meet and Teams calls is the part we are rolling out next; the story above is where it is going.
Let your agents sit in.
