Job card · start here

AI systems from the workshop floor.

The AI Fixer is where I document the machines, agents and experiments I build between real jobs. Useful beats impressive, evidence beats theatre, and anything with power needs a switch.

Live bench

Experiments

Agent colonies, local memory, kill switches, witness models and other work still being tested.

Series

Series

Guided topic hubs for agent safety, local-first control and prompt/spec discipline.

Downloads

Kits

Public-safe checklists and templates you can copy without exposing the private bench.

Notes

Notes

Field reports and longer thinking on agents, context, safety, local AI and business automation.

HOUSE RULES

What belongs here

  • working artifacts, not vague claims
  • operator judgement, not lab theatre
  • local-first control where the risk earns it
  • agents that can be inspected, stopped and corrected

What does not

  • private customer or family details
  • internal IPs, credentials or hostnames
  • fake certainty about unfinished experiments
  • automation without a human-owned boundary

Recommended first path

  1. The System Prompt Is Not the Safety System
  2. Why I Rebuilt My AI Agent’s Memory
  3. The Four-Machine Test
  4. Practical Agent Safety series
  5. Public-safe kits
  6. Bench Partner notes