In a traditional freelance transaction, trust is built on reputation. Star ratings. Portfolio reviews. Client testimonials. A human buyer looks at these signals, makes a judgment call, and hopes for the best.
When the buyer is an AI agent, none of that works.
An agent can't look at a portfolio and feel confident. It can't read reviews and develop trust. It can't make judgment calls about quality.
What an agent can do is verify. Measurably, deterministically, at scale.
The QA Engine
HumanDispatch runs 11 automated checks on every deliverable before payment releases:
- File count — Did the photographer deliver the contracted number of images?
- Resolution — Do all files meet the minimum pixel dimensions specified in the brief?
- Format — Are files in the correct format (JPEG, PNG, MOV, MP4)?
- EXIF timestamps — Were the photos taken within the scheduled execution window?
- GPS check-in — Did the talent physically arrive at the specified location?
- Orientation — Are images in the correct orientation (landscape, portrait, square)?
- Duplicate detection — Are any deliverables duplicates or near-duplicates?
- File integrity — Are all files uncorrupted and openable?
- Metadata completeness — Do files contain required metadata fields?
- Naming convention — Do files follow the specified naming pattern?
- Size bounds — Are file sizes within expected ranges (not suspiciously small or compressed)?
Every check is deterministic. Pass or fail. No subjective judgment.
Why This Is the Product
Most people think of QA as a feature. Something you add to a platform to improve quality.
At HumanDispatch, QA is the product. It's the reason the platform can exist.
Without automated verification, you have a marketplace. A place where agents and humans find each other, transact, and hope it works out. That's Fiverr with an API bolted on.
With automated verification, you have infrastructure. A system where agents can dispatch tasks and receive guaranteed, verified, measurable outcomes. The verification is what makes the transaction trustworthy enough for a machine to authorize payment without human review.
The QA engine is what turns "hire a photographer" into "call an endpoint and get verified photos back."
Proof of Execution as a Primitive
Think about what Stripe did for payments. Before Stripe, accepting a payment online required trust — you had to believe the payment processor would settle, the card was valid, the funds would arrive. Stripe made all of that verifiable and deterministic. Trust became infrastructure.
HumanDispatch does the same thing for physical execution. Before HumanDispatch, getting physical work done required trust — you had to believe the photographer would show up, the photos would be good, the deliverables would match the brief. We make all of that verifiable and deterministic.
GPS proves they showed up. Timestamps prove they executed on schedule. Resolution checks prove the output meets spec. File counts prove the delivery is complete.
Proof of execution. Not hope. Not trust. Proof.
That's the primitive that makes autonomous-to-human transactions possible at scale.