You cannot inventory your way out of shadow agents — they are created faster than any catalogue is updated. What works is making the gate the thing agents must pass, so ungoverned ones surface the moment they try to do real work.
Source: 2025 industry surveys of enterprise AI agent adoption.
Any developer with an API key can stand up an agent in an afternoon. It gets a service account, it talks to production systems, and it appears in no register. By the time a quarterly review notices, it has been making decisions for months.
That is how you arrive at roughly 60% of enterprises scaling agents against about 4% with governance, and 97% reporting an identity incident. The controls are not missing because nobody cares — they are missing because inventory-first governance cannot win a race against provisioning.
Traditional discovery scans infrastructure. Agents are not infrastructure. They are API consumers that can live anywhere, including on a laptop and inside someone else's cloud.
Stop trying to enumerate every agent first. Put verification at the points that matter — the delegation boundary, the dispatch decision, the cross-org handoff — and an unregistered agent fails the moment it tries to do something consequential. Discovery becomes a by-product of enforcement.
Make the compliant path the cheap one. Registration is free, takes one call, needs no approval ticket, and gives the developer something they want: an identity their agent can use to be trusted by other teams. Governance that costs less than evasion gets adopted without a mandate.
Then query the graph. /v1/discover lists registered agents by capability and trust, per organization — a live register that maintains itself because agents have a reason to be in it.
An agent nobody registered fails the gate the first time it matters. That is a faster and more reliable detector than any asset scan.
One free call, no ticket, and the agent gets a portable reputation out of it. Developers register because it is useful, not because policy said so.
Discovery reflects reality because agents keep themselves listed to stay dispatchable.
Revocation is immediate at every verification point, without waiting for a policy to propagate through anyone's infrastructure.
Discovery is the inventory. It stays current because registration is the path of least resistance.
verified reflects the trust
threshold and account status. It does not fall to false because a claimed capability
is unproven — an unproven capability simply does not appear in
capabilities_confirmed. Gate on both fields.No, and any product claiming to reliably do that is overselling. Assay makes unregistered agents fail the checks that matter, which surfaces them through behaviour rather than inventory.
Put verification on a boundary they need to cross, and make registration free and instant. Adoption follows incentive. A mandate without a gate produces a spreadsheet; a gate produces a register.
Yes. Discovery filters by capability, platform and minimum trust, and agents carry an owner org. That gives you a live view of what is registered under your name.
Those are governed by your existing internal controls and are a much smaller problem. The risk concentrates where an agent acts with a party your controls do not reach — which is exactly where Assay sits.