Meta, Sierra pitch Personal Agent Protocol to govern how personal AI agents deal with business

Sierra and Meta along with a group that includes Genesys, Instinct, Rocket, Shopify, Stripe and Walmart launched Personal Agent Protocol to define how a personal AI agent should interact with a business.

The rise of personal AI agents to carry out a wide range of human delegated tasks, but there are also a lot of issues ahead. Authentication, visibility into what AI agents do to websites and APIs and security are all issues.

In a blog post, Sierra noted that consumers want speed, dependability and trust, brands want visibility and control and companies want efficiency and access. Personal Agent Protocol is designed to enable agents like Meta's Muse work with businesses to get things done.

• Personal AI assistants to proliferate and so will the governance headaches

The spec is really early, but this standard could develop quickly much like Model Context Protocol did. Stripe and Shopify are early partners and that makes sense along with Walmart since most of what personal AI agents will do boil down to shopping.

A few takeaways about Personal Agent Protocol:

• What caught my eye was Rocket's participation. Will a personal AI agent make a home purchase? Rocket is probably thinking more about dealing with personal AI agents to complete mortgage tasks delegated by a human. Nevertheless, enabling a personal AI agent to make the biggest purchase of your life will be fun to watch. I encourage all of you to try it and let me know how it goes.
• Genesys and Sierra are obviously on the front end of this for the customer experience and service angle. They really want the visibility angle.
• Walmart is looking for the retail hook. You could take out a lot of commerce friction dealing with personal AI agents. Shopify and Stripe have a similar hook. I do wonder how personal AI agents can juice sales and how disputes would play out.
• Meta's blog on Personal Agent Protocol notes that Muse applies two tests before acting. First, there's the one honest person test where Muse weighs whether a reasonable human would do something like create five accounts to redeem one coupon. The second test looks at what would happen if every agent did something and whether the system would hold up. These tests are fascinating, but likely to be bunk. At some point, Muse is going to say, 'I'll do it because I can." After all humans do it.

More:

• The great enterprise AI rewrite is starting
• How Capital One evaluates, governs its AI agents
• OpenAI's Dev Day features dots, Ultrafast, GPT-6 Sol, but can it woo the enterprise?
• Agentic AI deployments: What you should, and shouldn't do