Use Claude for UI work and Codex for server tasks
A developer split interface work and backend work between Claude and Codex after other agents failed on real projects.
- What they did
- They run paid access to Claude Ultra and ChatGPT/Codex Pro, along with a few local models for minor jobs such as reformatting text and generating speech. They rely on the standard CLI and desktop apps, then connect them with a custom tool they built. Over the last month, they have sent UI tasks to Claude, giving it lots of reference material, while Codex has handled the backend.
- How it went
- Codex is the steadier choice for finishing server-side work, but its UI sense is weak. Claude produces better designs, yet it often stalls before wrapping things up. OpenClaw and Hermes were abandoned after they broke down on actual work.
- Tips
- Claude needs strong examples if you want good design output. This setup also means paying for several top-tier subscriptions at the same time.
Try it yourself with Claude
In [project folder], build the backend for [feature], including [endpoints, database tables, auth, or other server needs]. Match the existing style, run the tests, and then write a brief handoff note in [file name] that lists the routes, data structures, and anything the frontend must expect. Before any irreversible step such as deleting, sending, submitting, or deploying, stop and ask for my approval. Finish only when the tests pass and the note is saved.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Make a first app with Replit and Claude
A commenter used Replit’s agent with Claude Sonnet to assemble an app in about five minutes, despite having no real coding experience.
Use a team chat agent to spin up client prototypes
A small software shop let Hermes watch team chat and connected it to their repo and server tools so it could draft prospect prototypes after approval.
Build a local app for agent output cleanup
They used Codex to create Mianotes, a local place to collect agent-made files and links.
Review a pull request with staged sub-agents
Use parallel checks and a validation pass to review a branch or PR before you decide what to fix.
Run several coding agents from one interface
A developer uses T3 Code to manage multiple coding agents on personal machines, a VPS, and a separate work setup for code changes and writeups.
Plan work in a PRD, then assign the build
A developer asks a stronger model for a plan, turns it into tasks, and routes implementation to lower-cost agents.