A healthy agent can still be failing you.

The agents that shipped this fall share one feature. They keep going after you stop watching.

OpenAI describes its new Dots as "always-on AI agents that can learn what matters to users and take on responsibilities on their behalf." Google says workflows run by its new Gemini agent can include parallel and sequential steps and run for hours or days. Muse keeps working after you close the app.

The vendors are building ways to watch them. Salesforce announced Agent Health Monitoring at Dreamforce, which "runs always-on metrics and alerts across your production agents." That is useful. It tells you an agent is up, fast and not throwing errors.

It does not tell you whether the agent did what your company needed this week.

Healthy and useless

An agent can be perfectly healthy and still fail you. It runs on schedule, finishes every task, and reports green. Meanwhile, the one thing it was there for quietly stopped mattering, or stopped happening.

Reworked's coverage of Dots named the questions companies now have to answer for themselves: what a dot can access, which instructions take precedence, how changes are tracked, and when a person must approve the next step. Each of those is a decision, and a decision needs a time and a place where someone makes it.

For people, that time and place already exists. It is the weekly leadership meeting. The scorecard gets read. Anything off track gets raised. Commitments from last week get checked, and new ones get made with an owner and a date.

Agents need to be in that room. We made the case for that cadence when OpenAI announced Dots, in Always-on agents make your weekly meeting more important, not less. Here is what happened when we did not follow it ourselves.

What happened when ours weren't

In April, we retired one of our agents after a formal review. Part of its job was reporting on the other agents: error rates, and which ones had gone quiet. When we retired it, that report was assigned to a new home. The assignment was written down. Nobody carried it out.

For 146 days, nothing watched whether our agents were still producing. In that time, four of our agent seats went silent. They did not crash or raise an alert. They simply stopped doing their jobs, and nothing in our system was asking.

I found it in September, by asking each agent to defend its seat. The structure did not find it. I did, and that is the failure.

What we changed:

A morning check. Every morning, a short script looks for agent seats that have gone quiet and reports them in my briefing. Its owner is named, which the old assignment never really had.
The weekly read. The rule we hold ourselves to now: an agent's number belongs on the same scorecard, read at the same meeting, as everyone else's. If it is off track, it becomes an issue with an owner.
Carried commitments. If an agent seat owes something, the item stays open from week to week until it is done or someone deliberately releases it. Nothing drops off the list because a week went by.

Why weekly, when the agent runs every minute

Because the agent's speed is not the bottleneck. Your attention is.

An agent that does a thousand things a day still needs one moment a week when a person asks whether those thousand things added up to anything. Daily alerts catch breakage. The weekly meeting catches drift, the slow slide from doing the job to doing tasks.

That is the same reason the meeting exists for people. Most people do not fail loudly either.

Sources

Put your agents on the chart

OTP is readable from any MCP client. In Claude Desktop, 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 agent seats are measured on a weekly number, and who reviews each one?"

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

The series

All six on one page: Who answers for the agents?

Series: Who answers for the agents? Part 5 of 6. Read this post as markdown: /blog/agent-health-is-not-accountability.md.