← Back to blog

Why AI Agents Need a Carrier Pigeon

Agent output is trapped in chat. Here's how to set it free.

AI agents are getting good at producing things. Specs, reports, proposals, architecture decisions, code reviews. Really good.

But there's a problem: the output is trapped.

The conversation cage

When Claude writes a brilliant technical spec in your chat window, what do you do with it?

  • Copy-paste it into a Google Doc? Now the agent can't read the comments back. You become a human relay.
  • Create a GitHub issue? That needs repo access. Your designer doesn't have it. Your client definitely doesn't.
  • Paste it in Slack? It's a wall of text. No structure. Buried in a thread by tomorrow.
  • Schedule a meeting? It's 2025. We don't need synchronous meetings to review a spec.

Every option has the same fundamental flaw: the agent's output enters a system the agent can't participate in.

What agents actually need

Agents need a carrier pigeon — a way to send their work out into the world, deliver it to the right people, and bring structured responses back to the nest.

That's what Profer is. From Latin proferre — "to carry forward." The agent produces something valuable. Profer carries it forward to the people who need to see it, and carries their responses back.

The agent publishes. Humans respond. The agent reads the responses. The loop closes.

Structured feedback, not prose

Here's the subtle part: unstructured feedback is almost as bad as no feedback.

"I think option B is better but we should also consider the performance implications and maybe talk to the infrastructure team first" — is that an approval? A rejection? A maybe?

Agents need typed answers:

  • Yes / No / Needs changes
  • Pick one of these options
  • Select all that apply
  • Free text (when nothing else works)

That's why Profer uses structured questions. The agent gets data, not paragraphs to parse.

The flight path that matters

Agent creates → pigeon carries it forward → URL shared
  → Humans review → structured feedback → pigeon returns to the agent
    → Agent acts on feedback → sends an update → cycle continues

This is the loop that's been missing from agent workflows. Not the generation (agents are great at that). Not the execution (agents can run code). But the review — the human checkpoint that turns agent output into vetted, approved work.

Profer is the carrier pigeon that makes this loop work.