If you use two coding agents, how do you split the work?
by•
I use Claude and Codex on the same codebase, and the handoff matters more to me than raw coding speed. A second agent can catch a mistake, but only if it checks the work independently.
My split: Claude writes the brief and reviews; Codex implements in its own branch, then Claude reruns the tests.
If you use two agents, how do you divide the work and decide a task is ready to ship?
17 views
p/runevr
p/producthunt
p/google-cloud-platform
p/reso
p/many-displays
p/slotify-2
p/airuncode
p/ai-context-flow
p/kodlo
p/aperture-7
p/lasso-11
p/raycast
p/dino-kids
p/kuc-kuc
p/supabase
p/jev
p/offiai
p/murror
p/dex-by-exmergo
p/htmldrag
Replies
One agent wrote the data pipeline and another built the summary view. The final report had a confident number nobody had actually measured, and it took a day to realize the first agent had inserted a placeholder. Keep them on isolated stacks, not sequential handoffs.
Mine is split by job. Claude Code writes and reviews all the code. Codex only makes images, through its built in image_gen, because there's no per image cost on my plan.
What made the image side reliable was quoting every label verbatim in the prompt. When I described the boxes loosely it made up its own text. Do you hand Codex the exact tests it has to pass, or only the brief?