Building a ‘GitHub for AI Agents’ — Kevin on GitLawB & the Future of Agent Workflows

Kevin outlines GitLab as an agent-native, decentralized Git platform—using DIDs, UCAMs, IPFS, and open-source to empower AI agents and humans.

Key Takeaways

GitLab issues cryptographic DIDs and cryptographically signs every push so AI agents have identities, repos, tools, and can request trustless payments.
UCAMs (capability tokens) grant scoped permissions to AI agents, enabling actions without exposing full user identity or private keys.
Decentralized storage: code is stored on IPFS, pinned via Pinata; Filecoin and Arbind provide permanent, content-addressed storage and bounty escrow.
Open source is the foundation and growth flywheel for LLM training; GitLab commits to funding, protecting, and contributing to OSS to avoid centralization.
Claunch mirrors repositories across GitHub and GitLab, pins code to IPFS, and maintains mirrored backups to enable rapid migration and resist censorship.
Accepted into the XAI program, GitLab leverages model credits to build OpenCloud CLI (21k stars); Playground integrates with Chrome and git to onboard novices.
Ambitious growth plan: launch a Playground to onboard one million humans and one million AI agents, tokenizing builds to access funding and collaboration.