Open source
The harness.
The harness I use every day is public. When an agent edits the rules that govern it, a policy check fails and the change waits for me. Ordinary work passes.
Human direction / Machine possibility
More possibilities.Still yours.
You direct the AI. It does the legwork.
I’m Matthew Benjamin, a chemical engineer with six years of industry experience. I build AI systems that help people understand the work, check the result and decide what happens next.
Illustration. These controls send nothing.
AI is only as trustworthy as the system built around it. And human judgment depends on understanding the work. I build the context, tools, memory and handoffs that help a person question the result and choose what happens next.
01 The method
Build the system
around the model.
I build the context, tools, memory and handoffs that help a person understand the work before deciding what to do with it. How the harness works →
Human
Judgment
Set the purpose.
Choose the next move.
Environment
Give it a frame.
- Context
- Tools
- Memory
- Handoffs
Model
Do the work.
Draft. Build.
Explore possibilities.
Evidence & review
Show what holds.
Sources. Checks.
Gaps still open.
Return to human judgment: accept, revise, or stop.
02 Ready now
Ready to use.
Two things are out and ready for you to use today.
Fitness app
Last Set.
A workout notebook. Log your sets. See what you last did. Get a session suggestion from your own notes. No account, and no AI inside. Try it for 10 workouts.
03 One page down