Labs
Where we test the ideas that are too early to fund.
Labs is the part of the studio that gets to be wrong. We run two-to-four week probes against questions that don't have a market yet - new interaction models, agentic workflows, hardware-software seams - and publish what we learn.
probes run to date
graduated into ventures
median probe length
Frontier probes
Short experiments against a single sharp hypothesis. Most fail on purpose; the survivors graduate to Studio.
Agent systems
Tool-using agents with real evals, traces, and a human in the loop where it counts.
Interface research
New input and output models for software - spatial, voice, ambient, generative.
Open notes
We write up findings in plain language so the whole portfolio compounds on what we learn.
How a labs engagement runs.
- 01
Question
Frame a hypothesis tight enough that a two-week probe can answer it.
- 02
Probe
Build the smallest artifact that produces a real signal, not a demo.
- 03
Read
Score it against the hypothesis. Kill, iterate, or graduate.
- 04
Publish
Write the note. The negative results are worth as much as the wins.