I measured what a coding-agent hand-off saves, and where it didn't help

Description and use

Hand‑off between agents in Friday saves tokens and speeds up bug fixes, though manual copy‑paste yields similar outcomes.

Problem

Does handing a task from one coding agent to another reduce token usage, exploration steps, and improve speed compared to starting over or copying the previous agent’s answer?

Outcome

Hand‑off consistently used fewer tokens (median 21% fewer), required fewer exploration steps (median 2 vs 4), and produced faster first edits in 12 of 14 tasks, while all fixes passed the project’s tests. Copy‑paste performed similarly to hand‑off but lacks the automated handover convenience.

Workflow

• 1. Claude Code identifies and explains a bug without making changes.
• 2. Codex receives the task via hand‑off with a one‑line note, or starts over, or receives a copy‑paste of Claude Code’s answer.
• 3. Codex attempts to fix the bug in a fresh project copy.
• 4. The project’s own tests are run to verify the fix.

Evidence

Against starting over, a hand-off used a median of 21% fewer tokens (fewer in 13 of 14 tasks).

Source

The next agent explored half as much before its first edit: a median of 2 steps against 4, fewer in 14 of 14.

Source

14 of 14 fixes passed the project's own tests, every way.

Source

Copying Claude Code's whole answer into Codex by hand did about as well: the same tokens, the same exploring, and in Express it finished faster.

Source