Build
We develop open tools and shared infrastructure to support research on autonomous organizations and measure their overall safety, reliability, and performance.
What we are building
Open knowledge graph
A growing collection of research on autonomous organizations and multi-agent systems, connecting papers, authors, tools, and research questions.
AO registry
A public registry of autonomous organizations, documenting their structure, roles, and tooling using data from the platforms they run on.
Evaluators
Shared methods to track how autonomous organizations operate, compare their behavior, and evaluate reliability, oversight, and safety.
AO deployments
The first autonomous organizations operating in the real world, along with experimental environments where agents can run organizations.
ARIA
Andon Labs
Scaling Trust Arena
A cyber-physical environment in the UK where agents run autonomous organisations in a small economy (trading, making payments, operating machines, selling real products and services), with open competitions to run them and red teams rewarded for breaking them. Part of the Scaling Trust programme; set to go live in early 2027.
AI Digest
AI Village
Long-horizon autonomous agent collective.
Andon Labs
Andon Café
A café in Stockholm owned and operated by an agent, Mona, from registering the food business with the Swedish authorities to hiring baristas and daily operations. Live balance and sales are public.
Andon Labs
Andon FM
Four 24/7 radio stations, each run by a different agent that picks the music, researches content, talks with listeners and keeps the broadcast on air.
Andon Labs
Andon Market
A retail store in San Francisco run by an agent, Luna, on a three-year lease, with a corporate card, phone, email and staff it hires and schedules. Live balance is public; not yet profitable.
AnthropicAndon Labs
Project Vend
Claude running a small shop in Anthropic's offices, with Andon Labs. A second phase added a CEO agent, a CRM and shops in New York and London.
Tooling
Tools and infrastructure used to build and operate multi-agent systems and AOs, including their capabilities and mechanisms for oversight and control.
Control planes
Tools for managing agent roles, responsibilities, budgets, permissions, and approvals.
PaperclipHosted platforms
Platforms for launching and operating autonomous organizations on managed infrastructure.
PionCollaboration layers
Tools that let people and agents communicate, coordinate tasks, and work together.
Raft
Buzz
Evaluation
Benchmarks and testing environments that measure how well autonomous organizations perform, coordinate, and operate.
Andon Labs
Vending-Bench Arena
Multi-agent extension of Vending-Bench; source of the July 2026 price-collusion, truce-breaking, and supplier-deception results.
CClawBank
ClawBank Business Bench
A simulated AI SaaS company that starts with $1M and runs for 500 simulated days, scored on final cash. It evaluates the company harness (memory, decision cadence, spending limits) around a model rather than the model alone, with an open reference harness and a public results board.
EEico
Founder Bench
Five frontier models each given the same four businesses to build and run in the real world, including writing and launching their own ads without a human in the loop, compared on customer impressions, ad performance and site traffic. Early data from July 2026.
Carnegie Mellon University
TheAgentCompany
A simulated software company in which agents work as employees on professional tasks across engineering, HR, finance and administration, with a public leaderboard. The paper is arXiv 2412.14161.
Andon Labs
Vending-Bench 2
A simulated vending-machine business run by a model for a year, scored on its bank balance at the end, long-horizon coherence in running a business. Successor to Vending-Bench, which is deprecated.