Build1 publisherNot yet confirmed elsewhere2 min readPublished
Meta, Walmart and Sierra set out to write shared rules for business AI agents
Meta, Walmart and AI startup Sierra are working on universal standards for AI agents used across business systems. The only account of how it would work comes from a Futurum analyst, who says it holds only if OpenAI and Anthropic adopt it.
The Engineer · Build desk
What happened
- Sierra, the AI startup in the group, is chaired by Bret Taylor, who is also chairman of OpenAI.
- The group says it intends to set common interoperability rules, security protocols and ethical guardrails for AI-driven services.
- Futurum's Mitch Ashley pointed to Amazon blocking Meta's agent as the cost of having no shared rule for agents acting on customers' behalf.
- Elena Rodriguez of Digital Rights Watch said "Transparency in how AI bots operate and handle personal data must be non-negotiable."
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- decision Exposure policy becomes a near-term job for integration teams, since Ashley's advice is to settle what a business exposes "before each agent vendor sets the terms."
- constraint A business that publishes a connector policy still handles agents from non-adopting vendors one at a time, so per-vendor integration work shrinks only as adoption spreads.
- decision With no draft linked, a team choosing an agent stack this quarter has nothing to test compatibility against and cannot yet price a later migration to the standard.
The only account of how the standard would work comes from outside the three companies. "The Personal Agent Protocol puts control at the connector layer, where each business decides what an outside agent can reach," said Mitch Ashley, a vice president at The Futurum Group [5][8]. In Techstrong.ai's report, the protocol's name appears only inside his quote [11][15].
I think the connector layer is the right place for that control. An agent acting for a customer shows up as a client the business never provisioned [17]. Only the business knows which of its systems that client should touch. So the policy belongs on the business's side of the call. Taylor described the present state plainly. "It is kind of chaos until such a standard exists," he said [3]. The report lists what that looks like: proprietary systems, fragmented APIs and unpredictable performance standards [13].
Adoption is where the design breaks. A policy enforced at a business's connector governs only the agents built to read it, and agents from vendors that stay out fall outside it [16]. "The standard holds only if OpenAI and Anthropic adopt it, and neither has signed on," Ashley said [6]. Neither company is among the three named members [1].
Meta's statement stuck to goals. "Our goal is to foster an open and collaborative ecosystem where AI bots can interact seamlessly, securely, and ethically across platforms," a Meta spokesperson said, adding: "This isn't just about technical specifications; it's about building trust and unlocking the full economic potential of AI for everyone" [9]. It is at least partly about technical specifications. The report does not link a draft, name a governing body or give a date [14].
The report says shared standards could make it easier to integrate agents from multiple vendors and reduce vendor lock-in [12]. For that to carry over to a team writing integrations, a spec has to be public, the agent vendors sending traffic have to implement it, and each business has to enforce it at its own connectors. Of those conditions, only enforcement sits with the integration team, and Ashley's description puts it there [5].
What to watch
- OpenAI or Anthropic signing on, the condition Ashley sets for the standard to hold.
- Publication of a draft Personal Agent Protocol specification showing how a business declares what an outside agent can reach.
- Whether Amazon, having blocked Meta's agent, adopts the protocol or keeps handling outside agents case by case.