From purchase orders to shipping labels, we read every document that proves parts were built correctly, verify them, catch what's wrong, and maintain permanent records that ship with the parts
Skilled engineers rarely have the hours a lot takes by hand, so the agents read every record while you watch and every number opens its evidence. The hours and minutes marked * are estimates for one lot.
I · The problem
A single brake pad hides 80+ certificates from two dozen suppliers, each in its own format and language
Skilled engineers have scarce time for stacks this size, and on a safety part the one that slips rarely gets a second try
A finished car is built from thousands of parts, each with a pile like this
Standards like IATF 16949, AS9100, ISO 9001 and FDA QMSR make these records mandatory, the rules only tighten, and the pile grows faster than anyone can read by hand, so the time to bring agents is now
In 2025, in the United States
A recall is a defect caught after the part already shipped, when every affected vehicle on the road has to be found and fixed
II · The floor
Every day, a quality team works the same pile of certificate PDFs, arriving by email or keyed into an ERP one at a time, each opened and checked by hand, and one brake pad alone carries 84 certificates, so across every part on every line the math stops working
FactoryFloor is the operating brain for the lot, one record with six agents reading every certificate before the call comes to you, and the backing plate looks clean enough that a rushed reviewer would pass it, but you are the reviewer now, so watch what the agents catch
Every step above left a verifiable trace.
III · The transformation
Eighty-four certificates per lot, each a different format and language, more than a person can reconcile by hand
FactoryFloor folds them into one normalized record, sealed with Cloud KMS and a Merkle proof, a portable file a buyer or auditor can verify without trusting you
IV · The proof
Sealed part by part, the paperwork stops being a pile of PDFs and becomes one record you can see through, all the way down your chain, not just your direct suppliers
This is not a diagram of what we plan to build. Every box is running code or a Google service you can check right now, and the floor demo you just used ran on exactly this system.
SequentialAgent → ParallelAgent → SequentialAgent: one LlmAgent, five deterministic steps.Architecture diagrams usually describe an aspiration. The diagram in the chapter head above is audited against the running system: every box maps to code or a managed service you can poke right now, and the demo on the floor, the machine view in the X-Ray and that diagram are three renderings of the same thing. Each row below is one box and its proof.
vertex-gemini-2.5-pro and per field confidence · agents/document_extraction.pyagents/*_validator.py, spec_comparison.py, risk_scoring.pyagenttrust/authorization.pyagentsarmy-prod.factoryfloor.audit_ledger; GET /api/package/{session_id} serves the packagePOST /api/watchesScreens have been designed for human eyes for thirty years; the next decade of software will be read mostly by agents. So we design for both and call it AX: every human decision here is also a machine-verifiable record an agent can read over A2A, the protocol Google's agent stack speaks, so FactoryFloor is ready to talk to other enterprise agents by design.
You reviewed the part and signed the decision with your name.
An agent reads that same decision as a signed ledger row it can fetch and check.
You explored the brake pad and its 84 certificates on the map.
An agent reads those same certificates as a Merkle tree, recomputing the root to prove nothing was changed.
You read this page, written for your eyes.
An agent reads a signed card and calls FactoryFloor on its own, with no human in the loop.
Priced per verified record, the one signed lot the agents produce from a pile of supplier certificates. Steady volume pays less per record.
For pilots and low volume. No commitment.
Includes X records, then $x per record
For steady, repeating volume.
For OEMs and multi-site supply chains.
Illustrative model. Final pricing is set with our first design partners.