← use casesCode review
AI agents are fast. Reviewing what they write is the bottleneck. A common critique of agent managers on HN applies here: "you're just converting typing time into reading time." That critique is fair for tools that make review harder. Pane's goal is to make review fast enough that the net time savings hold.
pane's approach
shorten the review loop
The agent lives in the same pane as the diff. Press ⌘D and the syntax-highlighted diff view opens without leaving the keyboard. Approve, iterate, or commit — all from the same window. No context-switching between terminal, browser, and git client.
iteration without context loss
If the agent's output has a problem, switch from the diff view back to the agent terminal in the same pane. Paste a follow-up prompt. The agent corrects. Press ⌘D again. The diff updates. Review again. The agent and the review state are always in sync because they live in the same place.
walkthrough
- → Step 1. The agent writes a change in pane X. You get a desktop notification when it finishes (or when it needs input).
- → Step 2. Press ⌘D to toggle the diff view. The file tree and syntax-highlighted side-by-side diff open immediately.
- → Step 3. Browse hunks with ⌘J / ⌘K. Review the changes file by file.
- → Step 4. If you spot a problem — a regression, an unintended refactor, a missed edge case — switch to the agent's terminal tab in the same pane. Paste a follow-up prompt describing the issue. The agent fixes it.
- → Step 5. Press ⌘D again. The diff refreshes. Re-review the affected files.
- → Step 6. When satisfied, press ⌘K to commit. Write a message. Push. Done.
how this differs from vs code review
VS Code's built-in source control diff is excellent — but the agent (running in a terminal elsewhere) and the diff view are in different applications. You review in VS Code, iterate in the terminal, go back to VS Code. Each switch costs attention. In Pane, the agent terminal and the diff view are both inside the same pane. The loop is: diff → feedback → diff. No application switch.
what to watch for
- → Scope creep. Agents routinely refactor adjacent code that was not in the task spec. The diff makes scope creep visible — look for changed files that were not in the original task. Ask the agent to revert them if you want a clean, scoped PR.
- → Test file omissions. Agents sometimes implement a feature without updating the corresponding test file. Check the diff for missing test changes on every feature pane.
- → Context window degradation. On long agent runs, output quality can drop as the context window fills. If the diff looks sloppier than expected, the agent may need a fresh session with a tighter context. Start a new pane for the remaining work.
- - - - - - - - - - - - - - - -
See also: Pane vs Conductor for a comparison of review workflows across tools.