AI agent for trades and field services teams
Crews in the field. Training decays fastest here. Cognition keeps the playbook in their pocket and knows when to remind them.
An AI operating stack for trades and field-services teams makes install checklists, safety protocols, and customer scripts available on a technician's phone via an AI agent — with CLO modeling which crew members are about to forget which procedure, so refreshers happen before a callback or an incident.
The problem
- Your senior tech knows how to coax a reluctant old boiler back to life. Your new tech calls them at 9pm.
- Safety protocols are trained annually and forgotten quarterly. The only reminder is an incident.
- Your CRM has the install record. Your techs have 12 handbooks. Nothing is retrievable in the truck.
A company like yours
Jamal dispatches 35 technicians across a metro area. Mean ramp time for a new tech is 4 months — mostly because tribal knowledge about customer sites, equipment quirks, and install idiosyncrasies lives in his senior techs' heads. When a senior is on a different job, the junior calls dispatch, who calls the senior, who walks them through it over the phone. Every call is 15 minutes of somebody's day.
Before vs after
- ~4 monthsTech ramp to independent calls
- ~6 hrsSenior tech hours per week answering phone
- ~8%Callback rate due to procedure misses
- ~5 weeksTech ramp to independent calls
- <1 hrSenior tech phone hours per week
- ~3%Callback rate
The stack, in plain English
For field services: Claude Code (or a mobile wrapper) as the interface, memory-mcp or mem0 holding install history, safety procedures, and equipment notes. ServiceTitan or Housecall Pro MCP bridges feeding live job data. CLO tracking per-tech retention on safety and procedural knowledge.
What day one looks like
FAQ
Do techs actually want to use an AI on their phone in the field?
The ones who already pull up YouTube in the truck do. The interface is a voice question; the answer is one sentence. It's faster than calling dispatch.
What about poor cell coverage on job sites?
Claude Code + mobile requires connectivity. For offline-first field work, we recommend a cached memory layer (letta supports local-first) + sync when back in coverage.
Can this integrate with ServiceTitan, Housecall Pro, or Jobber?
ServiceTitan has a public API; community MCP bridges exist. Housecall Pro and Jobber are similar. The stack is adapter-agnostic — any API becomes an MCP server with a ~100 LOC wrapper.
Will this reduce callbacks?
Yes, measurably — because procedures and install quirks are retrievable at the moment of truth, not when the tech gets home and tries to remember. CLO further reduces drift by surfacing the decaying concepts before they cause a miss.
What memory tool fits a field-services stack?
memory-mcp (simplest), mem0 (multi-tech scoping), or letta (agent-style persistent memory per tech). Start simple. See cognitionus.com/memory.