This was the week AI agents stopped waiting for a prompt — and the week everyone started asking who watches them while they work. OpenAI shipped always-on Dots, and four kinds of supervisor showed up alongside: auditors, a hardware watchdog, an analyst warning, and a $100 million academy.
1. Dots Makes Four: The Always-On Agent Is Now a Category
OpenAI shipped Dots at DevDay, agents that keep working around the clock on their own cloud computers. With Grok Bot, Muse, and the open-source Hermes Agent already out, always-on agents are now a product category, and the differences between them are your buying criteria.
2. The White House Accord Brings Auditors to the Frontier
Six technology leaders signed a voluntary pact with four layers of control: internal controls, an oversight team, an independent external auditor, and a board committee. No law backs it, but it is a structure enterprises already know how to buy against.
3. Nvidia Puts a Watchdog in the Silicon
Nvidia's Open Agent Safety Platform pairs a deny-by-default sandbox with a monitor that runs on separate hardware from the agent it watches. Containment is moving out of the agent and into the infrastructure beneath it.
4. Gartner: Most Vendor-Built Agents Won't Survive the Handover
Gartner predicts 70% of enterprises will abandon agentic AI built by vendor forward-deployed engineers by 2028, trapped by cost and unable to evolve it themselves. The fix starts in the contract, before the engagement begins.
5. Anthropic Bets $100M on Training the Supervisors
Claude Frontier Academy aims to train 10,000 engineers by the end of 2027 through a residency-style program. Look closely at who is in the first cohort, and at whose stack the skills are built on.
The Governance Angle
The Week Always-On Agents Got Supervisors
CIO Corner: Rent the Build, Own the Supervision
It is reasonable to rent the build. It is not reasonable to rent the supervision, because the party that understands the agent is the party that controls it. Six moves for this quarter.
The 6-Layer Stack
Amazon pledges $1 billion to data center communities, a $300 million chip-smuggling arrest, agents moving onto vendor clouds, and the mid-tier model becoming the workhorse.
Agent 101: Determinism vs. Sampling
Why the same agent can do something different on the same task, and why one good run proves very little. A vending machine and a chef explain it.
Agents went always-on this week; the organizations that come out ahead will keep the supervision in-house.
See you next week — still watching, still distilling.
— The Distilled AI Digest Team
