AI Weekly: Who Owns the Skill

Who Owns the Skill

Datadog's security team built an LLM judge that reads the intent behind a piece of code, originally to scale review of third-party contributions. It turned out to catch injected code in real supply chain hijacks, then turned out to work on markdown. So now it sits in front of every agent skill entering the company, and what it finds is the news: "we put this judge in front of every skill that wants to get introduced. We actually find quite a bit of malicious skills in all these marketplaces."

That is Emilio Escobar, chief information security officer at Datadog, where more than 4,000 engineers work with coding agents. A quarter ago a skill was a text file on somebody's laptop. This week, in three unconnected conversations on three continents, it was a corporate asset with a gatekeeper, a budget line, and no named owner.

Escobar's structural observation is the one worth keeping. Permissions used to hold because only people who could write SQL could reach the data. "AI is going to find a way to get it. All you have to do is prompt it. It flattens the organization." His answer is role-based access at the tool layer and credentials that exist only for the length of a call, so an agent never touches a secret sitting on disk.

Singapore supplied the ownership question. Zen Foo, chief of staff at the fintech Aspire, argues the job lands on him by elimination: org-wide skills, governance, and policy are infrastructure, and only someone with whole-company visibility can hold it. He is honest about where his own firm sits, which is stage two of three: "we're at this stage where it's a bit of like beautiful chaos and we're just letting it happen." His warning to anyone reaching for a mandate: "All this stuff doesn't work if people aren't excited about AI." And on measurement he wants the obvious metric killed before it sets: "it has to be more than just like, how much tokens you're using?"

ServiceNow's Josh Newman gives that instinct a name. Fluency and tool adoption are trackable and disconnected from return, so he separates adoption from absorption and keeps readiness data advisory. "The goal is to use readiness as a compass," not an automated decision-maker about people.

The quietest version came from a fifteen-year-old communications agency. Moxie COO Kate Connor puts AI projects on the same published roadmap as every other strategic objective, for a reason that is not about efficiency: "what I don't want to happen is people build their own perception or their own idea of what they think is going to happen with AI." Her rule is that none of it can sit at the top. "AI can't just sit with me and the leadership team. It has to be brought up through the team."

Last week the agents got HR language. This week the artifact got a filing cabinet, a gatekeeper, and a budget line. Nobody has named the owner.

Sources: Emilio Escobar (Datadog), a16z, "The CISO Playbook for AI Agents | Datadog," Aug 11, 2026. Zen Foo (Aspire), Chief of Staff Network, "Why the Chief of Staff Should Own AI Adoption, with Zen Foo," Aug 12, 2026. Josh Newman (ServiceNow), HR Leaders, "How to Get Your Team to Actually Use AI at Work," Aug 12, 2026. Kate Connor (Moxie Communications Group), The Modern People Leader, "320 - How Moxie's COO built an AI strategy for a people business: Kate Connor, COO, Moxie," Aug 10, 2026.