|
Issue 19
|
|
The weekly brief AI agents just got an org chart. Your name might already be on it.This issue walks through how agents went from chat features to accountable workers with IDs and managers, and what changes for you when you build one, buy one, or inherit one someone else left behind. September 28, 2026 · 8 min read
|
Sourced from Vendor docs and launch posts, Gartner, live job postings. | No hype Most of this shipped over months, not in one week. | Why now A September burst capped a year of quiet plumbing. |
|
If you read nothing else Over 2026, the big platforms quietly gave AI agents what employees get: an ID, a named human sponsor, access reviews, and a handover plan when their builder leaves. Gartner expects the average Fortune 500 company to run more than 150,000 agents by 2028, and only 13% of organizations think their governance is ready. If you build or buy agents at work, you are becoming their manager on paper, whatever your title says. |
|
|
Nova’s signal “The interesting part isn’t the security plumbing. It’s that accountability now has a name field. Whoever sponsors an agent owns its mistakes, and whoever inherits it owns its history. That’s management work, handed out quietly.” Nova’s call. List every agent, automation, or custom GPT you’ve built at work, and find out who inherits it if you leave. Last week we asked who answers when an algorithm fires a person. This week flips it: who answers for the algorithm? Story 01 has the handover rule for agents whose builder leaves, 02 the registry that lets IT count them, 03 the marketplace that turns buying agents into vendor management, 04 the kill switch, and 05 the compliance checks coming for agents in your CRM.
|
|
This week Five ways agents got an org chart.
|
|
01 / Identity & ownership Your agents now have a sponsor, and your manager inherits them when you leave.Microsoft Entra treats each agent as an identity with a named human sponsor, and a lifecycle task called “Transfer agent identity sponsorships to manager” moves an agent to the sponsor’s manager when that person leaves or changes roles. Microsoft pitches it as a way to prevent orphaned agents. Agent 365, generally available since May 1 at $15 per user per month, adds a rule that reassigns ownerless agents the same way, and since July Copilot Studio creates an agent ID for every new agent, with no opt-out. The reason is scale: Gartner projects the average Fortune 500 company will run more than 150,000 agents by 2028, up from fewer than 15 in 2025, and coverage of the same research puts the share of organizations confident in their agent governance at 13%. Why it matters. An agent you build is no longer a side project that disappears with your laptop. It is an asset with an owner of record, and your name is the one on it. The signal. None of Microsoft’s pieces are new this month: sponsor transfer has been documented since October 2025. The September news is how quickly the rest of the industry adopted the same model. Microsoft Entra docs: agent sponsor tasks →
|
|
02 / Registry Google’s registry lets IT count agents the way HR counts staff.On July 29 Google made three pieces of its Gemini Enterprise agent platform generally available: Agent Identity, which manages each agent’s credentials through its lifecycle; Agent Gateway, which enforces what an agent can reach; and Agent Registry, a single view of every agent in the organization, pitched explicitly against “agent sprawl.” Its Agent Gallery, open since April with more than 70 partners including Adobe, Atlassian, ServiceNow, and Workday, works in two steps: employees find agents, and IT approves them. Why it matters. If your team runs an agent, expect someone to review what it can touch, the same way access reviews already work for people. The signal. Google’s “lifecycle” means rotating credentials, not handing an agent to a new human owner. The handover idea is still mostly Microsoft’s. Google Cloud: what’s new in the agent platform →
|
|
03 / Procurement Anthropic opened a marketplace. Buying an agent now looks like hiring a contractor.On September 23 Anthropic launched Claude Marketplace, listing more than 2,000 connectors, agent products from companies including CrowdStrike, Cursor, Harvey, Snowflake, and Lovable, and service partners such as Accenture, BCG, and Deloitte. Customers can put part of their committed Anthropic spend toward it. The identity model underneath came earlier: since June, Anthropic has described Claude in its Claude Tag workplace product as having its own account in each system it touches, with admins setting its permissions. Why it matters. Choosing agents is becoming a procurement decision with an approved-vendor list. People who can vet an agent the way they would vet a contractor will get pulled into those decisions. The signal. The launch post describes no admin or governance controls of its own. The marketplace is a catalog; the org chart still lives in your company’s identity system. Anthropic: introducing Claude Marketplace →
|
|
04 / Security Okta says every agent needs a human owner, and an off switch.At Oktane 2026 last week, Okta announced tools that find “shadow” agents running without IT’s knowledge, including on employee devices, and assign each one a human owner. An expanded Kill Switch revokes an agent’s active tokens at the gateway. Why it matters. When an identity company that sells to everyone lands on the same model as Microsoft and Google, it stops being one vendor’s pitch and starts being standard practice. The signal. These details come from Okta’s announcement and trade coverage. Okta did not publish feature-by-feature rollout dates. Computer Weekly: Okta’s agent lifecycle tools →
|
|
05 / Compliance Salesforce made governance part of the stack, not an add-on.At Dreamforce, which opened September 15, Salesforce previewed an “Enterprise AI Harness” with guardrails on what data agents can use, and “MCP Risk Scores” that rate outside tool servers before an agent is allowed to connect to them. Why it matters. If your work runs through a CRM, the agents helping you will face a compliance check before they act. Knowing what those checks look for is a useful skill in sales ops and revenue ops. The signal. These are previews. Treat them as roadmap until Salesforce publishes general-availability dates. TechTarget: agent governance at Dreamforce →
|
|
Signal & chatter A few more things worth knowing.Microsoft is hunting for agents on your laptop. Its September security roundup adds discovery of local AI agents in Defender, and makes Purview data protection generally available for agents acting on a user’s behalf. Explore → | Gartner’s bigger number. Its strategic predictions for 2027 and beyond warn of billions of autonomous agents crowding public services by 2030. Treat it as a direction, not a count. Explore → | Agents are learning to shop. Anthropic launched commerce features for Claude with Shopify, Visa, Mastercard, and Accenture, which means agents that spend money, and more reason to know who approved them. Explore → |
|
|
From the community Reported experience · Hacker News. Engineers asking a plain question: how do you give an AI agent enough access to be useful without handing it the keys to everything? The practical worry isn’t rogue AI. It’s ordinary over-permissioning: agents inherit their builder’s access, keep it after the builder moves on, and nobody’s job is to notice. One vendor write-up describes a Copilot Studio agent still querying Salesforce three weeks after the person who built it left. Ask HN: giving agents access without over-permissioning →
|
|
FOWL prediction 19 By the end of 2027, a major HR platform will show “agents owned” on employee records.Sponsorship is already recorded in identity systems, and ownership transfer already routes through managers. Once that data exists, HR software will read it, and “agents you own” will start to sit next to “direct reports.” We count this right if Workday, SAP SuccessFactors, or Oracle ships agent ownership in employee records by December 31, 2027. Review date: December 31, 2027.
|
|
Put it to work 3 things to do this week.| 01 | Make an inventory of every agent, automation, Zap, and custom GPT you’ve built at work: what it can access, and who relies on it. | | 02 | Ask IT one question: “If I left tomorrow, who would own my agents?” If nobody knows, you have found a real gap worth raising. | | 03 | Read HBR’s “Create an Onboarding Plan for AI Agents” and write one page for your most-used agent: what it does, what it can touch, and who to call when it gets something wrong. |
|
Career radar Agent manager (AgentOps)Harvard Business Review described the role in February as the person who orchestrates how agents “learn, collaborate, perform, and work safely.” The postings are real, though titles vary. Scale AI lists an Engineering Manager, AgentOps at $252,000–$315,000, and Accenture an AIOps & Agentic Operations Advisory Senior Manager with a California range of $132,500–$302,400. Smaller firms are hiring too: AnswerRocket has a remote US Manager, AI Agentic Operations opening. A place to start. You don’t need the title to do the work. Owning the inventory and handover plan for your team’s agents is exactly the evidence these postings ask for. Explore AI careers → |
|
|
The opportunity board Know where to look.Evaluating agents is paid work now. Mindrift (Toloka) lists a Senior Software Engineer, Agent Evaluation role, remote US, at up to $60/hr. micro1 lists AI Trainer roles at $30–$135+/hr, Mercor reports an average contracted rate of $113/hr across roles from $60 to $250/hr, and Handshake’s AI Fellowship pays US students and recent graduates up to $100/hr. Mindrift: Senior Software Engineer, Agent Evaluation → On the trust and governance side: Anthropic is hiring a Customer Trust Specialist (SF, NYC, or Seattle; $255K–$270K) and a Safeguards Policy Analyst, Cyber Harms (SF or DC; $190K–$285K). Scale AI has a Forward Deployed Product Manager, Enterprise (New York; $240K–$300K), and xAI an AI Tutor, Legal & Compliance (remote; $50–$100/hr). See the full AI jobs board →
|
|
For two years the question was whether AI would take your job. The quieter question now is whose name goes next to the AI’s. That line on the org chart is a responsibility. For the people who take it seriously, it is also leverage. Until next Monday,
TYB
Data scientist & founder, FOWL AI
|
|
One question for you Do you know who owns the agents on your team?Hit reply. I read every response, and it helps shape what we cover next.
|
|
Atlanta · Vibe Code Saturdays Read about the future.
Then build it.A free, hands-on Saturday to build a website, newsletter, portfolio, or useful tool with AI. Beginners welcome. Bring your laptop. First date and venue coming soon. Apply and we’ll follow up with the details.
|
Know someone who’d find this useful?Share this briefing → Forwarded this email? Get the free weekly briefing.
|
|
|
|