AI agents for field service with a fail-closed write-gate

Description and use

Foreman+ uses an AI agent fleet to capture equipment data from photos and voice notes, validates facts through a fail‑closed write‑gate, and stores only verified information, providing transparency and trust for subsequent technicians.

Problem

Field service technicians often lose or misremember key facts such as equipment model, serial number, and quoted price, leading to inconsistent records and re‑diagnosis by subsequent technicians.

Outcome

Foreman+ demonstrates that an AI agent fleet can reliably capture and store field‑service facts with a fail‑closed write‑gate, ensuring that only verified data is persisted while rejected proposals remain visible as claims.

Workflow

• 1. Technician takes a photo of the equipment nameplate and speaks a brief description of the problem.
• 2. The intake agent extracts model and serial number from the image; if unreadable, the field remains UNKNOWN. 3. The intake agent records the spoken notes as plain files. 4. The estimator agent searches the fleet memory for similar past jobs and proposes an estimate. 5. The closer agent builds a close‑out document without LLM calls and briefs the next technician. 6. Every proposed fact passes through the write‑gate, which performs identity, size, and Gemini verifier checks. 7. Approved facts are written to Cloud SQL; rejected proposals are logged with the verifier’s reason. 8. The ledger displays all writes, including refusals, with source chips and agent provenance.

Evidence

Foreman+ is an AI agent fleet for field service work. A technician photographs an equipment nameplate and describes the problem out loud, from a phone or from camera glasses.

Source

Between every agent and the database sits the write‑gate. A proposed fact goes through cheap checks first: is the proposing agent a registered fleet member, does the value fit a 4,096‑byte cap, does the subject stay within 64 predicates. Only then does a Gemini verifier judge it against what's already known about that subject.

Source

In the run captured for our submission gallery, the fleet made 14 proposals in 83 seconds. The gate approved 13 and refused 1: the 2022 date, with the verifier's reason quoted on screen.

Source