Companies as Code

Self-operating enterprises driven by a single source of truth and a cascade of events, with humans in the loop only where it matters, are no longer speculative. They're probable.

In 2025-2026, the stack for this has converged: agent frameworks (CrewAI, LangGraph, AutoGen), workflow engines (Temporal, Inngest), and LLM-native tool use have made it possible to encode company processes as executable graphs. The "codebase" isn't just the product, it's hiring flows, support triage, compliance checks, and go-to-market motions. Human triggers sit at decision points: approve, reject, escalate, or let the system run.

Research from agentic workflow papers and production deployments (e.g. AI-native ops at scale) suggests the bottleneck has shifted from "can we automate it?" to "where do we need a human signature?" Companies that get that boundary right will look like code: versioned, auditable, and iterable.

This note is a living draft. More to come as the research solidifies.