MIT professor says agent rules in a platform are hard to change
Pierre Azoulay of MIT Sloan says a platform must set agent access, agent identity and liability before launch. Changes after launch are costly and risky.
Pierre Azoulay, a professor at the MIT Sloan School of Management in the US, says AI agents make platform design harder. He spoke in a webinar for a four day course that he teaches with John Horton. The core rules of a platform are set before the first user joins. A change after launch breaks the deal that brought users in.
What he advises
Azoulay says a platform must select three things at the start:
- Which agents can enter: its own, approved ones or any assistant.
- How agents prove who they are.
- Who is liable when an agent transaction goes wrong.
He says a lean startup method, with a minimum platform that changes often, does not work here.
What agents change
Agents can make search faster and settlement cheaper. They can also add new costs: errors that spread between agents, and proof that a user is human. Azoulay says most data moats are illusions, because anyone can copy public data.
What is not known
The source gives no data or test. These are the views of one professor, who also teaches the course.
Insight: Platforms select their agent rules at this time
Azoulay says the agent access rule is hard to change. Spotify, which has 788 million monthly users, wants to be the music service that agents select. Adobe put its marketing agent into AI platforms from Anthropic, Databricks, Google, OpenAI and Microsoft. This reads as a race: the first rules that agents use can become the standard that others must follow. Leaders who wait can get rules that others set.
Sources
Posted