Event-Driven Background Agents for Coding
Zach Lloyd, CEO of Warp, on the 3 phases of AI coding
This points to the real endgame in AI coding being event driven automation, not chat based pair programming. The valuable agent is the one that wakes up when a build fails, a crash hits production, or a ticket lands in the queue, then investigates, edits code, and hands back a diff without a developer sitting there. That shifts the product from a better coding interface to infrastructure for wiring agents into the software delivery loop.
-
The split is between an interactive agent and a background worker. Interactive tools like Warp and Claude Code start when a developer opens the tool and prompts it. Automated agents run from CI, a production machine, or a webhook, and use the same repo and tool context without any UI.
-
Two product models are emerging. Devin is framed as an AI teammate that takes on whole software tasks. Warp is pushing the toolkit model, where developers program smaller agents into existing workflows, like bug triage, dead code cleanup, lint fixes, or first pass incident investigation.
-
That makes the integration surface the key battleground. Warp is centered on the terminal and CLI because it already touches Git, Docker, filesystems, and external tools, while it can also pull in shared MCP context. The more agent logic sits inside those existing pipes, the less important the traditional IDE becomes.
Over the next phase, coding tools will look less like editors and more like control planes for fleets of narrow software agents. The winners will be the products that can plug into source control, CI, issue tracking, and observability systems, then reliably turn those events into safe code changes and reviewable outputs.