Agent-in-the-Loop
Let agents check each other's work.
Sometimes the right second opinion isn't human. Phrony lets one agent review, challenge, or refine another's output — with the full handoff recorded.
Approve claim. Coverage valid. Fraud score 0.18.
Agreed on coverage. Flagging history: 3 claims in 90 days. Recommend additional review.
The problem
Humans are the right reviewers for the decisions that need judgment. But for the decisions that need a second pass on accuracy, speed, or consistency, routing every case to a person is slow and expensive. You need a scalable way to add a second set of eyes.
What Phrony does
Phrony lets you designate reviewer agents — specialists whose job is to check another agent's output before it's finalized. A drafting agent proposes. A reviewing agent verifies. If they disagree, the case escalates. Every handoff lives inside the same case record.
AITL · for scale
~ 1.4sEvery case gets a review pass. No queue, no waiting.
HITL · for judgment
~ 9 minFor the calls that actually need a person. Slower, on purpose.
Use AITL for scale. HITL for judgment. Most teams use both.
What you get
- Scalable second opinions. Every case can get a review pass, not just the flagged ones.
- Specialists reviewing specialists. A fraud agent verifies a claims decision. A compliance agent verifies a support response.
- Structured disagreement. When two agents don't agree, you see exactly where and why.
- Escalates cleanly. Disagreement can route to a human reviewer — HITL kicks in from there.
- Draft & Review
A drafts. B verifies before publish.
- Parallel Verify
Two agents reason in parallel. Compare.
- Challenger Model
A new model challenges the production one.