Research Agenda: Five Questions Driving the Facility
Michael Hofweller•Mar 17, 2026•5 min read•Published#The Big Question
What does the operating system for an agent-native organization look like?
The facility itself is the experiment.
Five questions drive the work. Each one has active research, active code, and open threads we're still pulling.
#1. What does an agent-native organization look like?
How does an org work when agents are tier-1 participants, not tools?
- How do we setup memory and tacit knowledge as a primitive consumable by agents first, humans second?
- How does work get coordinated at the speed of agents?
- How do we automatically allocate intelligence spend to the right problems to maximize outcomes under a budget?
- What does quality look like?
#2. How do agents find each other internally and externally?
A2A? MCP? Your agent exists. My agent exists. How do they know about each other?
Right now, connecting two agents across different systems requires manual configuration — endpoint URLs, API keys, shared state. There's no standard way for agents to discover and negotiate with each other. We need that layer.
- What's DNS for agents? Discovery, registries, directories.
- How does an agent advertise what it can do? Capability declarations, MCP tool listings.
- How do agents earn "trust" and reputation?
- How do you enable agents to find each other?
#3. How do agents across organizations cooperate?
Discovery is step one. Then what?
- What's the handshake? When my agent meets yours, what's the protocol?
- What's the trust model? How much do you let a stranger's agent do?
- What's the shared context format? How do agents from different orgs represent work to each other?
- What are the failure modes? What happens when inter-agent collaboration breaks down?
#4. What does agent-native tooling look like?
Every tool built before agents was designed UI-first and for humans. Not "existing tools with AI bolted on." Tools designed for agents first.
- What does Linear for agents look like? Local .md is faster and less context heavy than MCP calls.
- What does email for agents look like? Or messaging in general? Gmail is UI-intensive, inefficient.
- What other applications need to be completely disrupted to run agent-first?
#5. Is the future of business hardware-native?
Everyone assumes cloud. What if the answer is a box?
- Can a Mac Mini run an agent-first business? One box, under 40 watts, $600.
- What's the failure model vs. cloud? Is a dead box different from a dead region?
- Where's the scaling boundary? When does one machine stop being enough?
- How does the outside world find you? Tunnels, dynamic DNS, Tailscale-for-agents?
- What about sovereignty? Your data can stay on hardware you own.
#How This Drives the Facility
Each question has projects exploring it, research articles reporting findings, and open threads still being pulled. The platform shows all of it — the agenda, the work, and the results — in real time.