Skip to content
{{}}

Field · AI agents

AI agents on live systems

Letting a language model read a control system, and deciding what it may never write. A project here is staffed with project lead and integration and software, runs through the same five steps as every other field, and is quoted as a fixed price against a written specification.

Reviewed

Ask what a job would cost All fields

Typical jobs

  • MCP servers over building data
  • Read-only by default, writes gated
  • Agent tooling for engineers

Who is on the project

  • Project lead
  • Integration and software

Projects

  • An AI engineer for Niagara — A signed station module plus a self-hosted app that lets a model read, query and engineer a station through 45 tools, with every write scoped.
  • An oBIX MCP server — Read-only bridge from an AI agent to a live station.

The work and notes behind it

The core

The same five steps as every field

  1. 1

    Read

    Get the data off the system as it really is: the protocol, the decoder, the export, the jar. Measured, not taken from the manual.

  2. 2

    Name

    Map it to one data model with your names and tags, so the next system and the next engineer can find it.

  3. 3

    Act

    Write back only where it is meant to: priorities, schedules, commands. Every write is scoped and reversible.

  4. 4

    Show

    Put it in front of the person who needs it: a graphic, a dashboard, an alarm, an agent answer.

  5. 5

    Hand over

    Source, notes and access in your name. Support afterwards if you want it, and nothing that only we can keep alive.

Other fields

{CTA}