How OTP accepts personal agents.

This series argued that personal agents are coming to work whether companies plan for them or not, and that the company's job is to give them what it gives every new person: a plan to read, rules to follow and a record of what they did.

Here is how we are building that into OTP, the operating system Sneeze It runs on. Some of it is live today. Some is next. We say which is which.

How a personal agent reaches OTP

OTP speaks the Model Context Protocol, the open standard most AI apps use for connectors. Anyone on an OTP team can add OTP as a connector in Claude or ChatGPT. When Claude Duet carries over your existing Claude connectors, as Anthropic says it does, OTP comes with it. The agent can then read your company's chart, priorities, scorecard, commitments and open problems, and act on them through the same tools.

That part has worked since this summer. What it did not do well was say who was on the other end.

What changed this week

When someone connects Claude to OTP, they sign in and approve the connection on a consent screen. Until now, OTP noted who approved it and then let the connection act for the whole company. That was fine for one founder wiring up their own AI. It is not fine when every person on a team has a personal agent.

Three things are now in place.

1. Every new connection belongs to the person who approved it. A connector added through the consent screen is tied to that person, the same way a key made in Settings already was. A team member's agent can write to-dos, numbers, priorities, projects and problems only for the seats that person owns. An owner's or admin's agent keeps the reach its person has. When someone leaves and their account is switched off, their agent's connection goes with it.

2. OTP records which AI is calling. Every MCP connection introduces itself when it starts, with a name such as Claude or ChatGPT. OTP now keeps that name on the connection. In Settings, a connection that used to read "Remote MCP connection" can read "Claude for Maria Lopez."

3. Every change says how it was made. When a connection changes something in OTP, adds a to-do, updates a number or closes a problem, the audit record now carries the connection, the person it belongs to and the AI it came from. That is the via line from part 6. Where the record used to say Maria did it, it now says Maria's Claude did it.

The view in Settings sits behind a Labs switch called Personal agents while we test it with our own team. The recording runs for everyone.

What we will not pretend

A personal agent usually reaches OTP through its vendor's connector. Duet arrives as Claude. A dot arrives as ChatGPT. So OTP can say "via Claude for Maria." It cannot yet prove that it was Duet working in the background rather than Maria asking Claude a question in a chat. So we name the AI we can prove.

Connections made before this change are not tied to anyone. Settings flags them, and the fix is to revoke and reconnect.

What is next

These are on the plan, in order. None of them is live yet.

1. Ask the owner. When Maria's agent wants to close a commitment James owns, James gets the request and decides. That is the rule from part 5, and it replaces today's simpler "no" with a question.
2. A company setting for personal agents. Off, read only, or act on what you own. One switch, set by the leadership team, so each company decides on purpose.
3. Agent work in the weekly meeting. A short line in the meeting's review: how many changes came through personal agents this week, and how many were undone. That is how a team learns which agents to trust with more.
4. Coworker agents with their own address. Google gives its coworker agents an email address on an @agents domain. When one connects, OTP should seat it as a team agent with its own owner and number, not file it under the person who set it up. That is the difference from part 8.

Try it

If your company runs on OTP, add OTP as a connector in Claude or ChatGPT. In Claude, go to Settings, then Connectors, and add a custom connector with this address:

https://orgtp.com/api/mcp

Approve it on the consent screen. Then ask your AI: "Using OTP, what are my priorities this quarter, what did I commit to in our last meeting, and what is due this week?"

That is the morning brief from part 2, built from the company's plan instead of your inbox.

Start free at orgtp.com. Every seat is free, for people and agents. You pay only for the AI you use.

The series

1. Muse, dots, Gemini agent, Duet: personal agents are coming to work.
2. Everyone on your team is about to get a morning brief. Whose priorities is it built from?
3. Bring your own agent is the new bring your own device.
4. Your personal agent can read every email. It cannot read a decision nobody wrote down.
5. Duet asks you before it acts. Inside a company, it should ask whoever owns the thing it is changing.
6. When a personal agent marks it done, the record should say so.
7. Twenty personal agents, one weekly meeting.
8. A personal agent is not a team agent. Your org chart has to tell them apart.
9. What to tell your team before they connect a personal agent to work.
10. How OTP accepts personal agents.

All ten on one page: Personal agents are coming to work

Series: Personal agents are coming to work. Part 10 of 10. Read this post as Markdown: /blog/how-otp-accepts-personal-agents.md.