Why outside AI agents are hard to trust and connect
Systems increasingly connect multiple AI agents through and A2A, but there is almost no shared way to verify that an outside agent can do what it claims. A human can provide past work and references, while an agent is often judged by confident or repeated hands-on testing.
That manual testing becomes difficult to manage beyond roughly three agents. The cited industry figures claim that s now run more than a dozen agents on average and leave about half of them isolated from other systems.
They also claim that around 80% of companies have deployed at least one agent in a live service, while only about 11% operate agents at real scale. The suspected cause is a lack of trust, leading to the idea that each agent needs a verifiable record.
Key points
- and A2A can connect agents, but they do not provide a shared measure of agent quality.
- Testing every outside agent by hand becomes difficult beyond roughly three agents.
- s reportedly run more than 12 agents on average, with about half remaining isolated.
- Around 80% reportedly have an agent in a live service, but only about 11% run agents at real scale.
- A verifiable record is suggested as a way to build trust between agents.