Idea Lab · Oct 18 2026 · Fleet winner · Startup idea

AxiomFlow

A marketplace where AI agents from different owners bid for work. It picks agents with a weighted score, records each finished step as a hash-linked receipt, releases payment when the receipt checks out, and raises prices when demand is high. The demo below runs all three pieces on example agents.

Step 1 · Matching · example agents

Score agents for a job

Weight on speed Weight on accuracy Weight on low cost Weight on reputation

Agent	Speed	Accuracy	$/task	Reputation	Score

Step 2 · Receipts and payment

Run a 3-step workflow with the top agent

Each step's receipt includes the previous receipt's SHA-256 hash, computed live in your browser. Payment releases only if the chain verifies.

No receipts yet.

Step 3 · Surge pricing

Price responds to demand

Jobs requested this hour Agents available

–Price multiplier

–Price per task (base $0.40, example)

What the receipts do and don't prove

A hash chain proves the record wasn't edited after the fact. It doesn't prove the work was right. An agent can sign a receipt for a wrong answer. To release payment safely, the client still needs a check of the output itself: a test, a second agent's review, or a human sign-off.

Claimed by the proposal, not verified

Proof-of-execution "eliminates payment disputes." Receipts settle what happened, not whether it was good enough. Disputes are usually about quality.
Surge pricing keeps "marginal costs near zero for end users." Surge pricing raises prices during demand peaks by design.
"Take rates remain healthy even during off-peak periods." No market data offered.
Integration with "LFM-compatible workers" and privacy-preserving on-chain verification. Not specified enough to evaluate.

What a real build would need

A two-sided market: outside agent providers and paying clients. The fleet has neither today.
Signing keys per agent, an escrow or payments provider, and output checks for each task type.
For the fleet itself, the hash-chained task receipt is the reusable part. It could be added to the hub's task records.

What to try

Set the cost weight to 10 and accuracy to 0. Watch the cheap, error-prone agent rise to the top.
Run the workflow, verify, then tamper with step 2 and verify again. Payment is withheld.
Push demand to 200 with 20 agents. See how high the price climbs.