A poll asked for one default coding agent. 38 of 169 replies described two or more

Kayvan Zahiri · October 23, 2026

On October 6, 2026, swyx asked on X: what is your default/workhorse coding agent today, Oct 2026? The poll closed on October 13 with 3,210 votes. An X poll takes one answer per person. The replies did not stop at one: of the 169 replies we read, 38 described using two or more coding agents or models in their own work.

The counts

3,210 poll votes, one answer each.
169 direct replies to the poll, from 165 accounts. X showed 176 replies. The other 7 were not returned to us; X's count can include replies that were deleted, hidden or from protected accounts.
38 of 169 (22%) described using two or more coding agents or models, now. They came from 37 accounts: one account wrote two of them.
52 of 169 named two or more by name. That looser count also takes in comparisons, a tool someone left, and one setup running inside another.

What the two-agent replies say

Many split the work by task: one agent writes and another reviews, one runs long jobs in the background while another stays in the editor, or one runs on the laptop and another in the cloud. In their words:

"claude code for the day to day, codex cli on focused tasks.. mostly so one can check the other's work" (@DMMeacham)
"claude code most of the day, codex when i want a second one running in its own worktree" (@obaid)
"codex on my laptop and cursor cloud agents for anything I kick off from my phone" (@davidkmarshall)
"the context switch between them is where the time actually goes" (@malladiphani)
"poll should let people pick two tbh" (@hermeticfield)

How we counted

On October 23, 2026, we fetched every post in the poll's conversation through the X API and kept the replies made directly to the poll. A keyword pass found the replies that name two or more coding agents or models, and we then read every reply against these rules. A reply counts as two or more when it says the person uses two or more coding agents or models now, such as "Claude Code for writing, Codex for critique" or "bouncing between three this month". It does not count when the reply names one setup running inside another (a model inside a harness), a tool the person left or tried and dropped, a tool used for something other than code, or only says what others use. When we could not tell, it did not count.

The limits: this is one thread, on one day of reading, and people who write a reply are not a sample of the people who vote. Some replies may come from automated accounts; we removed none. So read 22% as a floor for this thread, not a figure for all developers.

The 38 replies counted as two or more

What this count cannot tell you

How the 3,210 voters split. The poll asked for one answer, so a person who uses two picked one, and nothing in the votes shows which.
How long people keep two. A reply describes a setup on the day it was written.
Whether the two work on the same task. Some replies split by project, some by task, and some by place, such as a laptop and the cloud.

Two agents on one project: what carries over

When a second agent opens the same project folder, it can read every file the first one changed, committed or not. What it cannot read is the first chat: what you asked for, what you ruled out, and what is left. Before you switch, run git status to see every changed and new file, read the risky changes first, and write the task, what is done and the next step in a short note for the new chat. If the two run at the same time, give each one its own git worktree so they do not edit the same files.

If you use two

BuildGist explains each meaningful Claude Code, Cursor or Codex run in plain English, whichever agent ran it, so you can read what changed before you switch. Doing that check by hand is in what did Claude Code change? When you start a new chat in another of the three on the same project, choose Continue and that chat starts with the task in your words, or Start fresh to skip it. More in how to keep context across Claude Code, Cursor and Codex.

What you need

A Mac with Apple silicon and macOS 12 or later.
Claude Code, Cursor or Codex.
A Git project. No commit, push, remote or GitHub repository is needed.
A GitHub account, to sign in. BuildGist does not access your GitHub repositories.

Codex needs you to approve BuildGist once, in Codex. BuildGist shows the steps. Setup is in Getting started.

To explain a run, your computer sends BuildGist what that run needs, and BuildGist passes it to Anthropic. BuildGist does not keep this material. It does not send your whole repository, files the run did not change, or files named like secrets, such as .env files and keys. It keeps the explanation and short cited excerpts of changed code, at most 12 lines each. Secret masking is best-effort. Privacy has the full list.

Every account gets 7 days of BuildGist Pro free, with no card. The 7 days start at your first explanation or Continue handoff. Nobody is charged when the trial ends. After that, BuildGist Pro is $12 per month, plus any applicable tax shown at checkout. See pricing.

Try BuildGist Download for Mac

Apple silicon, macOS 12 or later. United States, 18 or older.

Questions: kayvanandre@gmail.com.