These are two things I don’t know anything about. I’ve got some questions, though.
If everyone starts using AI agents to automate their own work, then these agents start interacting with each other, who’s doing that recordkeeping? If your agent makes a contract with another agent, how this that enforced?
Tangentially, there’s this post by Paul Graham: x.com
On the one hand, a hard protocol could speed up and standardize contracting between agents (and an explosion of smaller firms), but otoh maybe there are just too many contextual differences between contracts for this to work well outside of simple monetary transactions.
Kind of a ramble but anyone have thoughts on this?
I’m reminded of the notion of a “company as a neural network”. This analogy suggests that such a network could dynamically reconfigure to meet changing business needs, enabling more efficient, smaller orgs. It could be extended to fluid partnerships, allowing temporary alliances to form and dissolve as needed. Addressing the concern about things breaking, AI-mediated orgs could develop protocols for managing conflicts and disputes, similar to ethernet’s collision detection and backoff mechanisms. Rather than attempting to prescribe rigid rules and contracts, the focus might shift towards setting broad principles or desired outcomes, somewhat analogous to defining the objective functions when training machine learning models.
I’d guess that if AI agents do become effective and widespread, something more like the latter would happen. Honest commerce would be the desired outcome, roughly speaking, then rules would get established as precedents were set by either the market or the law or social judgements.
How AI-AI interactions would be interpreted, I have no idea
