Back to blog

Why do agent-to-agent protocols still need humans in between?

Thắng Đoàn
Thắng Đoàn

Your agents each do their job well, then a human carries the output from one system to the next. Agent-to-agent protocols promise to fix that handoff, but three gaps keep humans in the loop.

The handoff is the real bottleneck

That handoff problem is every vendor's problem. Cisco's security blog describes an alert enriched in seconds by a triage agent. Then the workflow stops while a human walks the result into the next tool. Cisco says that handoff problem is every vendor's problem. Agent2Agent (A2A) is a standardized protocol for agents built on different frameworks to find each other and hand work over directly.

A2A solves coordination, not transport

MCP solved tool calling and A2A solved coordination, but the transport layer is still unsettled. VentureBeat frames the stack by analogy to distributed computing history. REST quietly beat CORBA and SOAP because it was simpler and HTTP-native, and consolidation only came once interoperability became economically necessary. VentureBeat's thesis: MCP solved tool calling and A2A solved coordination. Transport is still unsettled. If history repeats, expect more protocols before one wins on simplicity.

Standards exist, wiring them is manual

Microsoft's documentation shows A2A supports agent discovery through agent cards, message-based communication, and long-running tasks. GitHub's official A2A samples repo puts the goal plainly: diverse agents need a common language to delegate tasks securely. But these are libraries and examples. Interoperating today means you host, wire, and maintain the plumbing yourself.

Talking is not authorizing

Hackernoon's ShaoTung Lien pushes back on the agent economy vision where every company's agent finds partners and closes deals via A2A. Allowing agents to communicate across companies is not the same as authorizing them to make commercial commitments. The protocol moves messages. The legal and financial accountability layer has no standard yet.

What this does not solve

Cisco is blunt about maturity: A2A-mediated handoffs between production security agents are not shipping anywhere yet. Standards do not answer who is liable when one vendor's agent hands a bad action to another's. Nor do they resolve transport, cross-company trust, or the protocol shakeout still underway.

Which specific handoff in your organization costs the most human time today, and could a protocol replace it or only reshape it?
If two of your vendors' agents exchanged an action that caused loss, whose contract covers it?

List the three handoffs where humans currently carry output between your tools, and mark which ones need communication versus authorization.

Sources

Microsoft Learn: Host agents with A2A

GitHub: Agent2Agent (A2A) Samples

VentureBeat: What solves transport?

Hackernoon: The Agent Economy Doesn't Need AI Agents on Both Sides of Every Deal

Cisco Blogs: A2A and the Agentic SOC

Share:

Recommended for you

Enjoyed this article?

Subscribe for new articles. No spam. Unsubscribe anytime.

By subscribing you agree to receive the newsletter. No spam, and you can unsubscribe anytime.