Use Codex to move research work faster
A theoretical chemist divided research tasks across models to speed up simulation-heavy work, while keeping a final check for tougher decisions.
- What they did
- The researcher used ChatGPT and Codex throughout a workflow focused on simulations of molecules, excited states, and spin effects. They arranged the setup so one model handled most of the analysis, a second model coordinated the process and did the last review, and a lower-cost model took care of repetitive tasks like starting and stopping jobs, moving files, and extracting data. For harder calls, they chose when to bring in more capable reasoning.
- How it went
- They said this setup let them accomplish in one day what would normally take a week, and that over the last year it enabled more research than they had done in the prior 10-plus years. More recently, their weekly allotment was exhausted in a day; OpenAI later said a bug skewed their usage and issued $500 in credits, and the researcher said they had also gone through $3,000 in credits quickly before.
- Tips
- A split workflow can work well if you reserve the strongest model for the parts that actually need it. Track spend against output closely, because the same setup can become too costly to sustain.
Try it yourself with Codex
I’m investigating [scientific question] with the code and data in [folder path]. Have [model A] handle the main analysis, [model B] act as the coordinator and final reviewer, and [model C] do repetitive tasks like launching runs, stopping them, moving files, and parsing outputs. Before any irreversible step such as submitting, deleting, or paying for a run, stop and ask for my approval. If you hit a step that needs deeper reasoning, flag it and wait for guidance. Finish with a short report of the results, any assumptions, and the next analysis to try.
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
Build a robot arm CAD package from a single request
A text-to-CAD setup turned one prompt into a multi-part robot arm model with motion logic, a viewer, and export files.
Repair a formal proof by iterating on code and builds
Codex fixed CertiGC proofs in Rocq with human review on any change that could alter the specification.
Inspect a laptop BIOS dump for hidden menu checks
A Reddit user had Claude Code analyze a BIOS dump, patch the signature gate, and expose hidden firmware settings after flashing the edited file.
Scan code for business-logic and auth bypass flaws
A developer used Devin to split a codebase into parts, trace exploit paths across files, confirm issues in a sandbox, and open fix pull requests.
Use an agent to draft parametric geometry systems
A Rhino plugin let a coding agent turn a brief into editable geometry logic stored alongside the project files.
Calibrate a robot hand and test gestures safely
Use an OpenClaw agent to inspect a 16-joint printed hand, verify motion with a camera, and calibrate firmware before trying simple gestures.