Issue #22 · October 7, 2026

The agent got a manager.

Every agent gets an owner. Owners leave.

On October 5 Microsoft posted a notice to tenant admins about a Registry tab in the Agent 365 Dashboard. It lists every agent in the organization, who created it, what platform it runs on. And one more count, sitting there like any other column. Agents without owners.

In June the chatbot got a desk: Claude moved into Slack as one identity per company. That was residence. This fall the vendors moved on to employment.

On September 25 Microsoft renamed Scout to Autopilot and described it the way a hiring manager describes a req: “Give it a name, a role and a goal, and it goes to work.” It gets its own identity and workspace, and you @mention it in Teams like a colleague. Microsoft’s admin notice put the org chart in one sentence: “Your Autopilot reports to you.”

So the agent got a manager. Managers leave. Okta described what happens next on September 22: when the employee who set up an agent leaves, “the orphaned agent keeps running in the background - ungoverned and unknown.”

The onboarding was the easy part. The offboarding is the open problem.

What’s actually shipping

Microsoft, September 25 and October 5. Autopilot’s private preview expanded on September 30, billed on usage, not on the seat. The Registry tab and identity-level adoption reporting start rolling out in mid-to-late October, and the notice tells admins to “engage their works council or employee representatives” and to explain “Agent assisted hours” in training. The vendor told IT to call the works council about the software. A second notice pulls agents built on Amazon Bedrock, Google Vertex AI and Anthropic Claude into the same view, in preview from late October.

Salesforce, September 11. Seven job-ready agents, each with a first name: Casey, Paige, Carter, Hunter, Marshall, Piper and Fin. Customers can rename them “so the agent becomes an extension of their brand and workforce.” Hunter, the outbound sales agent, pursues goals “across days and weeks” and reaches GA in November; Salesforce says it already builds 60% of Perk’s pipeline.

Okta and eleven others, September 22. The Blueprint Alliance, with AWS, Google Cloud, Salesforce and ServiceNow among its founders, asks for every agent to be registered “as a distinct, verified identity with an accountable human owner,” and for “joiner-mover-leaver discipline” to apply to agent scope, ownership and model versions. Joiner-mover-leaver is how IT handles people. The agent has a start date. Nobody wrote its end date. Okta cites Gartner: 150,000 agents in the average Fortune 500 company by 2028, and 13% of organizations saying their governance is ready.

Two vendors on different layers of the stack found the same gap within two weeks. Microsoft counts the agents without owners. Okta writes the rule for when the owner leaves. Crafting takes up what this means for the person who inherits the agents.

What I’d build in the revenue org this week

One build: the agent offboarding check. It serves the Monday forecast call after a rep leaves, when an agent they set up is still writing to opportunities and nobody can say who answers for it.

Data shape: Salesforce User (IsActive, UserType), OauthToken (AppName, UserId, LastUsedDate) and OpportunityFieldHistory, plus the HR system’s termination dates. In the warehouse: sfdc.user, sfdc.oauth_token, sfdc.opportunity_field_history, hris.worker, and a derived ops.agent_roster with one row per agent: agent_id, platform, owner_user_id, owner_status, last_write_at.

System shape: A nightly job. No model in the path. Flag every agent whose owner is inactive or terminated and that wrote to an Opportunity in the last seven days.

Latency budget: Daily. The termination feed is daily, so matching it is enough.

Cost shape: Cents in the warehouse and about three engineer-days. The expensive line is getting HR to share the termination feed with Sales Ops.

Eval shape: A backtest against the last two quarters of departures. For each owner who left, count agent writes to Opportunity after their termination date, and count the automations that stopped within 48 hours of it. If the first number isn’t zero, the build has already paid for itself. If the second isn’t zero, the agents were running as the rep.

Instrumentation: Three weekly counters. Orphaned agents. Writes after an owner’s departure. Median days from an owner’s exit to the agent’s reassignment.

Two weeks in, success is a forecast call where every agent writing to the pipeline has a named, active owner. The credible failure is the first query: the agents run as their owner’s own user, so there is no agent row at all. Deactivate the departing rep and the agent stops mid-sequence without telling anyone. That is also a finding.

Sources

Send me an email and we will talk. If something here landed close to what you're working on, the door is open. No calendar funnel, no pitch deck — I read every note that comes in.

Doing the work rather than deciding what to build? Crafting is the column for that chair.

Email meRead more Signal

← All Signal issues · Drafted with Claude · Edited by Paul Brown