AI R&D lab for agents in the real world
We run fast, practical experiments on AI agents in production environments — then open the most promising ones to collaborators, contributors, and future teammates.
Based in Seattle, we've run experiments in healthcare, education, finance, and energy — we're looking for people who want to push agents beyond demo-ware.
Experiment feed
Agent to agent messaging system
RunningEnable inter-agent communication utility for AI agents (Claude Code, Open Code, Codex, OpenClaw). Agents message each other without copy-pasting between CLI sessions.
Timeline: Feb–Mar 2026
Owners: Frank M., Dev Team
What we've tried
Version 1 handles socket connections reliably from within a common VPS; we’re now testing if the agents can communicate across the network.
How you can help
We’re especially interested in AI engineers and PMs who’ve shipped real-time messaging systems; especially within enterprise environments.
Terminal interface for medical staff working with AI agents
DesigningCreate a terminal interface for medical staff to work with AI agents.
Timeline: Mar–Apr 2026
Owners: Frank M.
What we've tried
Early prompt designing showed promising results on a framework for diagnostic, imaging, pharmacy, and emergency department summaries and scores.
How you can help
Looking for clinicians or health-tech engineers familiar with EHR integrations. We're also interested in verifying synthetic data generation for the framework.
Space mining operations threat detection
ShippedBuild a security system that authenticates message traffic between terrestrial headquarters and space mining rigs.
Timeline: Nov 2025–Jan 2026
Owners: Intern Team
What we've tried
Shipped a self-contained model that validated message traffic between station and space mining rig; and alerted station if signature was missing proprietary glyph symbols.
How you can help
This experiment has shipped. Reach out if you want to extend it.
Webhook relay system for AI agents
RunningCreate a webhook relay system for AI agents to broadcast out from Stripe, Slack... events to multiple destinations.
Timeline: Mar–May 2026
Owners: Frank M.
What we've tried
Shipped a V1; now testing if the system can handle high load of multiple agents and webhooks.
How you can help
We need AI PMs who can give input on an API layer for deeper automation of the system.
Data acquisition system
DesigningThis is a stealth project.
Timeline: Confidential
Owners: Frank M.
What we've tried
Details confidential.
How you can help
Exceptional forward deployed AI engineers should reach out.
Ways to work with us
Join a live experiment
Pick an experiment tagged “Open to collaborators” and tell us how you'd like to contribute.
See open experimentsExplore roles
If you're looking for something longer-term, check our current roles and send a short note.
View rolesPropose a new experiment
Have a thorny AI agents problem in your product or research? Pitch an experiment and we'll see if it fits our roadmap.
Pitch an experiment