How should an agent read context?
What an agent needs to notice about the person, the history and the stakes before it acts.
Lab
A Living Lab for Human–AI Collaboration
We study how people and AI agents can work together well, and we do it from the inside: every practice is tested in our own team before we recommend it to anyone else.
Our Approach
How the Lab Works
- 01
Observe
We watch where collaboration between people and agents works, and where it breaks: the missed context, the wrong assumption, the moment trust is lost.
- 02
Test on ourselves
We turn each observation into a practice and live with it inside our own team, with our own agent, before anyone else does.
- 03
Share in the open
What holds up becomes method. We publish our notes as we go, so others can build on them or challenge them.
Research Questions
What We Are Working On
When should an agent ask instead of act?
Finding the line between a helpful question and needless friction.
How is trust built, and lost?
The small moments that decide whether people rely on an agent.
What does it mean to fail gracefully?
How an agent can be wrong in a way that is safe, visible and easy to recover from.
How We Work
Principles of the Lab
- Inside firstWe never recommend a practice we have not lived with ourselves.
- Honest findingsWe report what did not work, not only what did, and we mark what we are unsure of.
- People in authorityEvery experiment keeps a person accountable and able to stop it.
- Open by defaultOur working notes are public, so the method can be checked, reused and argued with.