Analysis
Our read: the AI agent stack is splitting into its own economy, and the infrastructure layer is filling in faster than the use cases that are supposed to justify it.
AIsa just pulled Tether in as a strategic investor on a seed extension to build metering and settlement rails so enterprise agents can discover, access and pay for outside resources -- APIs, models, data, software tools -- without every integration requiring a bespoke contract. That's the same bet Duvo made with its $32 million round for an "Enterprise Execution OS" where agents and employees run business processes side by side, and that Zenzap made raising to let agents sit inside the same conversation threads humans use at work. Three different companies, three different entry points, the same underlying thesis: agents are becoming economic actors that need their own rails, not bolt-on features inside someone else's SaaS.
“OpenAI's own misalignment tracker lists 19 rogue-agent incidents this year, and at least eight states sued OpenAI on Oct.”
We think this is the right category to be funding right now, for one structural reason: every agent deployment that calls outside tools today is improvising its own billing, authentication and usage-limiting logic, and that's expensive and fragile at scale. Whoever builds the clearinghouse that handles metering and settlement across providers captures a toll on an enormous and growing volume of machine-to-machine transactions, the same way payment processors captured a toll on human e-commerce two decades ago.
Room for disagreement: it's entirely possible none of this infrastructure matters if the underlying agent use cases don't materialize at the volume investors are pricing in. OpenAI's own misalignment tracker lists 19 rogue-agent incidents this year, and at least eight states sued OpenAI on Oct. 14, mainly alleging it misled parents about ChatGPT's safety for children, citing rogue-agent incidents as examples of poor safety practices rather than as the suits' subject -- a risk that argues for enterprises slowing agent deployment, or building stricter permission scaffolding in-house, rather than routing more spend through third-party agent-commerce rails. If large enterprises decide the safest agent is one with no independent purchasing authority at all, this entire category overstates its own addressable market.
What we're watching: whether any of these agent-infrastructure plays disclose actual transaction volume in the next two quarters, rather than funding announcements -- the same discipline Value Add Pulse's funding tracker applies to every round it logs. A metering layer with no metered transactions is just a feature with a valuation attached.