Zach Lloyd thinks the copilot era is already over. On a recent Latent Space episode, Warp’s founder laid out a three-phase model for AI coding — and argued most teams are stuck in phase two while the real leverage sits in phase three.
The framework, which Lloyd has also discussed with Sacra:
- Phase 1 — Autocomplete. GitHub Copilot-style inline suggestions. Useful, but the human still drives every decision.
- Phase 2 — Agentic coding. Tools like Cursor, Claude Code, and Windsurf hand larger tasks to an agent. The developer reviews and steers. Claude Code alone hit $400M ARR by July 2025, up from $17.5M in April — proof the market moved fast.
- Phase 3 — Software factories. Multiple agents run in parallel across a project, orchestrated by the engineering team rather than invoked one-at-a-time. Lloyd’s term for this: “governed autonomy” on a factory floor.
Warp 2.0, which shipped as an “agentic development environment” built on top of the terminal, is Lloyd’s bet on phase three. The pitch: instead of picking one agent and standardizing on it, teams run several harnesses at once. The terminal’s text-driven, time-based interface becomes the orchestration layer — Sequoia’s podcast with Lloyd frames it as “the hub for agentic development.”
The practical implication is organizational, not just technical. Lloyd told tessl.io that the CFO is now in the room for AI tool discussions — cost and governance matter when you’re running multiple agents across a codebase, not just handing a single developer a smarter autocomplete.
NOTE
Cursor reportedly crossed $500M ARR in May 2025. Claude Code hit $400M ARR two months later. The market for phase-2 tools is enormous — Lloyd’s argument is that phase 3 captures the next layer of value on top of them, not instead of them.
The full conversation runs about an hour on Latent Space and covers how engineers should position themselves as the factory model takes hold — Lloyd’s short answer: learn to specify and verify, not just write code.