← Field Notes

Comparison · August 27, 2026

by Daniel Clark

Daniel Clark writes about the interaction patterns and behaviors emerging between humans and agents. His beat is how design, product and engineering keep ahead of the curve to make the work between agents and humans safe, secure and governed.

Codex shows you the plan. Claude Code makes you approve it.

The same transparency artifact — a step list — plays two different roles depending on whether execution waits for you.

Both flagship coding agents externalize a plan. The divergence is what the plan is for.

In Claude Code, the plan blocks: research first, present the steps, wait. The human edits or accepts, and only then does execution start. The plan is a contract, and deviations from it surface as updates against something you agreed to.

In Codex cloud runs, the plan narrates: a live step list that updates as the task proceeds. It's genuinely useful — you can see where the run is and what's left — but it authorizes nothing. Execution never waited for you.

Neither is wrong. A plan that waits costs latency and attention; a plan that narrates costs the person their say. But they're different patterns wearing the same UI, and calling both "planning" hides the choice a designer actually has to make: does the step list block, or narrate?

This is the kind of divergence the side-by-side on the pattern page is built to hold — same anatomy part, a different relationship between the person and the run.