Every AI now plugs into everything. Your operating plan should be one of the things it plugs into.

Yesterday OpenAI spent a good part of DevDay 2026 on one idea. The assistant should be able to reach the rest of your work.

Look at the list. MCP Events let connected apps trigger automations. Agents built on the Agents API can connect to MCP servers. Plugin extensions let developers put sidebar entries, interactive panels and file viewers inside ChatGPT and Codex, so a whole application feels native there. Sign in with ChatGPT launched with 16 partners. ChatGPT now answers when someone types @ChatGPT in Slack or Microsoft Teams, and it brings its connected tools with it. And dots, the new always-on agents, can be connected to over 4,000 apps, according to OpenAI's announcement as reported by 9to5Mac.

That is real progress, and it is the right direction. The plumbing problem is being solved.

Which means the next problem is not plumbing.

When every assistant can reach every app, the quality of what it does is no longer decided by the apps. It is decided upstream of them. It depends on whether the assistant can read what your company decided. What matters this quarter. Who owns what. What the scorecard says. What was promised in last week's meeting.

An agent with access to everything and no plan is a very fast intern with no manager.

This post walks through the three upstream causes of bad agent work I see most often, in our own company and in the companies that ask me about this. None of them is fixed by another connector. All of them are fixed by making the plan something the AI can plug into.

Cause one: it cannot see the priorities

The first cause is the simplest. The agent does not know what the company is trying to do this quarter.

Give an assistant your inbox, your CRM and your project board and ask it to help. It will help. It will clear the oldest items, answer the loudest threads and tidy the board. All of that is work. Very little of it is necessarily the work.

Priorities are the filter a good employee carries in their head. They know the three things the leadership team committed to this quarter, so they know which email matters and which one can wait until Friday. The assistant does not carry that filter unless someone hands it over. And in most companies, the priorities live in a slide deck from the planning day, a doc nobody opens, or the CEO's memory.

This is upstream of every app the agent touches. Connect it to 4,000 apps and it still does not know that this quarter is about retention, not new logos. It will be excellent at the wrong thing.

The fix is not a better prompt. The fix is priorities that live in a system the agent can read, the same system the leadership team reviews every week. When the priority changes in the meeting, it changes for the agent too.

Cause two: it cannot see who owns what

The second cause is ownership.

An assistant that can post in Slack, update a board and draft email will act on whatever signal reaches it. If it cannot see who owns that signal, it will either duplicate someone's work or step on it.

At Sneeze It, our chief of staff agent, Radar, runs the morning briefing. Early on, it pulled data directly from the sources. It scanned ad accounts itself, even though another seat owned ad performance. The result was two alerts for the same problem from two places, and a team that stopped trusting either.

The fix was not a new integration. It was ownership, written down. Every seat writes its own output. Radar reads what the other seats produced instead of scanning the sources itself. Once the chart said who owned ad performance, Radar stopped doing that job and started reading the owner's report.

That is the whole point. Access tells an agent what it can reach. The chart tells it what is its job. Only one of those is upstream.

Cause three: it cannot see what was already promised

The third cause is the one that costs the most, and it is the least visible.

Every week a leadership team makes commitments. Someone will call the client. Someone will fix the onboarding step. The CEO will make a decision by Thursday. Those promises are the actual operating state of the company. They are more current than any priority document.

An assistant that cannot see them will propose work that was already assigned, chase a problem already solved, or miss the one commitment that is now three days overdue.

Our scorecard seat, Tally, exists for a narrow version of this. It reads the sources behind each number and pushes the value to the scorecard. When a source is broken, it does not guess. It reports that the number is missing. That rule matters because the scorecard is what the meeting trusts. An agent that fills a gap with a plausible number is worse than one that leaves the gap.

Commitments need the same discipline. They have to live somewhere that is current, owned and readable, or the agent is working from last month's version of the company.

Where OTP fits

This is the layer OTP is built for.

OTP runs the weekly leadership meeting, which we call the Delta Meeting: agenda, scorecard, priorities, issues, to-dos and an Ollie Insight. Humans and AI agents sit on the same org chart and the same scorecard. To-dos are assigned to a person or a team, and they carry into the next meeting until they are done. The CEO's commitments get chased as loudly as anyone's.

The OTP MCP server puts all of that inside any MCP client, such as Claude Desktop, Cursor and others. Your chart, your scorecard, your priorities and your to-dos become something the assistant can read before it acts. And it works in the other direction. Agents claim a seat on the chart, push their own KPI and log their work back, so the meeting sees what the agents did the same way it sees what people did.

OpenAI is making it easy for an assistant to reach everything. OTP makes sure one of the things it reaches is the plan.

Frequently asked questions

What did OpenAI announce at DevDay 2026 about connecting apps?

At DevDay on September 29, 2026, OpenAI announced MCP Events for connected apps to trigger automations, plugin extensions that make apps feel native inside ChatGPT and Codex, Sign in with ChatGPT with 16 launch partners, ChatGPT in Slack and Microsoft Teams, and MCP server support in the Agents API.

Why does an AI agent need access to the company's operating plan?

Access to apps tells an agent what it can reach. The operating plan tells it what matters, who owns what and what was already promised. Without that, an agent does a lot of work that is not necessarily the right work.

How does OTP connect to AI assistants?

The OTP MCP server exposes your org chart, scorecard, priorities and to-dos to any MCP client. Agents can also claim a seat, push their own KPI and log their work back into OTP.

Sources

Plug your plan in

Our chart, scorecard and priorities are queryable from any MCP client. Yours can be too.

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 show me Sneeze It's chart. Which seat owns ad performance, and which seats read its report instead of pulling the data themselves?"

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

The series

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 5 of 5.