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

Arun Dass — Mon, 02 Nov 2026 04:38:15 +0000

I build Friday, an open-source app that runs coding agents side by side. One thing it does is hand a task from one agent to another, for example when one hits its usage limit. I wanted to know if that actually helps, or if it just sounds nice. So I measured it.

TL;DR

Against starting over, a hand-off used a median of 21% fewer tokens (fewer in 13 of 14 tasks).
The next agent explored half as much before its first edit: a median of 2 steps against 4, fewer in 14 of 14.
14 of 14 fixes passed the project's own tests, every way.
But pasting the first agent's whole answer in by hand did about as well. More on that below.

What I tested

Claude Code first found and explained a bug, and changed nothing. Then Codex fixed it three ways, each in its own fresh copy of the project, with one message from me:

Hand-off: Codex picks up the task with a one-line note.
Starting over: a fresh Codex session with the task restated.
Copy and paste: a fresh Codex session with Claude Code's whole answer pasted in.

I ran 14 tasks, 6 of them in Express. Everything was measured from the session log, not from what the tools say about themselves, and checked by running the project's own tests in the folder Codex worked in.

What held, against starting over

	Hand-off vs starting over
Tokens	Fewer in 13 of 14 tasks, a median of 21% (about 49,000 per hand-off). In Express alone, fewer in 6 of 6.
Exploring before the first edit	A median of 2 steps against 4, fewer in 14 of 14
Speed	First edit sooner in 12 of 14, finished sooner in 11 of 14, by a median of about 7 seconds
Fixed	14 of 14 every way. In Express, all 1,288 of its tests passed after every fix.

In Express alone the finish times were even, so I make no speed claim.

What didn't hold

Pasting 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. So I'm not claiming anything against a careful copy and paste.

What a hand-off gives you is that handover in one click, every time: what was asked, what was found, which files were read and what was run, without you being the clipboard.

The caveat

I ran this on the hosted version of Friday on 13 and 14 October 2026, before the open-source app existed. The open-source app has hand-off too, but I haven't run this benchmark on it, so these numbers aren't a measurement of it.

The data

Every run, the method and the raw data, including the runs I set aside, are on the benchmark page.

If you can find a flaw in how I measured, I'd like to hear it in the comments.

The open-source app is at github.com/splashxmoon/friday: MIT, no account, on your own computer.