HumanLayer is joining Factory and open-sourcing its multiplayer coding agent platform.
Dexter Horthy says agent swarms can now do months of work in days, making human alignment and control more critical than ever.
Multiplayer agent sessions force enterprises to solve audit, authentication and access when one agent acts across Slack, Notion and GitHub for several people.
HumanLayer joins Factory to bring multiplayer agents to big teams
HumanLayer, a multiplayer coding agent IDE, is joining Factory and open sourcing its platform. Its bet was that coding agents fail teams before the code exists, in planning and documents, so it made plans and code fully collaborative in real time and let anyone jump into a running agent. That same multiplayer setup creates the access and audit problems, because an agent acts for one person while a whole team watches.
Takeaways
Deal puts frontier tech on a stable platform. Horthy says the main reason he liked the Factory deal, struck after talks with Matan and Eno, is landing HumanLayer's frontier technology in a mature, stable product, and giving users things they had asked for over six to nine months.
Aligning on a plan before coding cuts rework. The slowest way to work with an agent is to ask for one thing and review it all at once. Customers who aligned on a high-level 200-line plan document first needed less rework and shipped with much greater velocity.
Multiplayer sessions create authentication and audit problems. An agent acting for one person may read a company's Slack and edit its documents. Once several people share a session, the system must compute their minimum shared access or assign an authentication profile, and attribute actions to a human upstream.
Sync engines are the hardest engineering problem. HumanLayer explored sync engines and conflict-free replicated data types, the approach behind collaborative editors like Google Docs and Figma. The system has to merge changes arriving late or twice and keep one source of truth across stateful components.
Swarms compress months of work into days. Because agent swarms can now do months of work in a couple of days, Horthy argues humans must align with them more than ever and stay the end decision makers, particularly in regulated industries where mistakes can mean millions in fines.
Context must move both ways around agents. Horthy says the higher-leverage work is pulling context and decisions out of the human before the run and pushing back what the agent did and what needs the most attention during review.
How HumanLayer customers move to Factory
HumanLayer customers are invited into Factory within a couple of weeks rather than the next few months.
Factory's operating principle is to earn the right, putting what users loved into the product before asking them to buy it.
The HumanLayer cloud keeps running for a set period so customers have time to transition.
The platform is open source, so anyone can run HumanLayer themselves.
Several larger customers have asked for help moving to self-hosted deployments.
Tap a name to go there · tap a line to see why they are linked