George Tsiolis

One habit I picked up earlier this year with coding agents is saving notes from some sessions inside each repository, but never committing them. Kinda like portable memory and context I can reuse with ANY coding agent in future sessions.

Everything gets singificantly harder when you have to convince the people instead of writing the code.

The remote playbook becomes the AI playbook. The patterns that reduced coordination costs in remote teams are now becoming the operating system for AI-native organizations. AI adoption is an organizational design problem, not a tooling problem. AI agents are remote collaborators. They're never in the room, never part of the side conversation, and never benefit from undocumented context. They work with artifacts your organization leaves behind: documents, issues, decisions, code. That’s why the operating principles by all-remote organizations like @GitLab matter more than ever. Documentation becomes executable context. Transparency becomes infrastructure. Decision velocity becomes the constraint. Async collaboration becomes loop design. Remote teams cut coordination cost with clear handoffs instead of standing meetings. Agents run the same way. The loop matters more than the prompt: what you assign, how you check it, where it hands back. Documentation becomes executable context. Docs, issues, and decisions are now runtime input for agents. Knowledge that isn't documented can't be retrieved. Knowledge retrieval replaces knowledge sharing. The author becomes the first reviewer. AI multiplies output, not review capacity. Self-review before requesting review becomes a required validation step. Transparency becomes infrastructure. AI can only reason over visible context. Hidden decisions become blind spots. Transparency is no longer about culture, it’s about reliable execution. Breaking the single source of truth creates execution bugs. Two conflicting docs create coordination costs, and make agents act with false confidence. Motion becomes cheap. AI makes producing work almost free. More PRs, more drafts, and more experiments don’t necessarily mean more progress. The real metric that survives is whether the work solved the problem. Bias for action becomes a permission model. Remote teams trusted people to act without waiting. Agents need the same trust, but with explicit boundaries. Reversible work should flow automatically, and irreversible work should pause for human judgment. Decision velocity becomes the competitive advantage. Every principle above reduces the coordination cost of making decisions. Organizations that optimize for decision velocity by making more informed decisions, faster and more often, will outperform those that simply produce more code, documents, or pull requests. The organizations that learned to coordinate distributed teams are best positioned to coordinate distributed intelligence. 🤖