Agent Engineering HandbookGitHub

Idea 04 of 19 The nineteen ideas

Preview environments matter more now that code is built outside your machine

On this page

Remote agent work makes previews more valuable

Theo says the old objection to previews was that everyone already built the application locally. Agents now build in cloud environments, background tabs, worktrees, and other machines beyond the developer's laptop. He argues that reachable previews matter more, including ways for agents to test them and post results.

Give agents previews they can reach and test

Plan a per-branch preview with a URL the agent can reach from its own workspace. Pair it with a scripted check of a key flow, screenshot, or console errors. Give the agent a way to attach the check results to the pull request when those actions are authorized.

Previews help when local review is impractical

This approach fits work where you cannot or should not run the application locally for every review. A preview gives the agent-built change a place to be inspected from outside your own machine. When local review already covers each change, this argument alone does not establish a need for preview deployments.

Previews are argued for, not measured

Theo presents an argument for previews, rather than a measured result from comparing review workflows. Setting them up can involve deployments, which remain outside the authorization of this writing task. The idea still applies to planning an agent's review environment, with deployment handled under its own authorization.

Watch at 05:06 · Home · Agent Engineering Handbook, September 9, 2026