Run a planner and cheap executor subagent coding pipeline
The commenter uses Codex with GPT-5.5 as planner and a GPT-5.4 Mini agent as executor so cheaper models handle tasks that would otherwise cost too much.
Done withCodex
- What they did
- In the thread, the commenter says they built their own workflow instead of relying on Codex’s default agent flow. They use GPT-5.5 only to plan tasks, then hand those tasks to a GPT-5.4 Mini agent that does most of the actual work, specifically including file editing; the person chose and assembled this routing themselves.
- How it went
- The claimed result is lower-cost handling for most coding work because GPT-5.4 Mini covers most tasks while the pricier model is reserved for planning. The comment gives no metrics, examples, or failure cases.
- Worth knowing
- The main catch is that they had to build the workflow themselves; the comment links a GitHub project but gives no setup time or cost numbers.
Try it yourself with Codex
In [repo path], set up a two-stage workflow for this task: [task description]. First, use the strongest model to write a step-by-step plan with small, well-specified subtasks and acceptance tests. Then hand each subtask to a cheaper, smaller model as an executor subagent, and have the planner review each result. Stop when all acceptance tests pass, and give me a summary of the plan, the changes, and any subtasks that needed rework.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: Reddit · Jul 26, 2026
Related use cases
Turn tickets into merged pull requests
Pulls tasks from GitHub Issues or Linear, then runs Claude Code or Codex in isolated Kubernetes pods to carry each one through to a PR.
Screen research papers against criteria
Codex reads a folder of PDFs to find articles that disagree with an argument, screens about fifty papers for a systematic review, and cleans messy CSV or Excel files.
Research Reddit customer pain points with a skill
The author replaced a Reddit research agent with a Codex skill that finds pain points, verifies communities, collects evidence, clusters problems, and analyzes commercial signals.
Build a C compiler with a team of parallel agents
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Turn customer feedback into tested pull requests
In OpenAI's own launch example, a Dot watches feedback for recurring requests, scopes small fixes, builds and tests them, and returns PRs with demo videos.
Attempt to build a navigation app (did not finish)
A poster tried to build a navigation app with Manus but spent a month of credits and over 8 hours without a working app.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.