20 real things people did with Codex, with the source and a prompt to try it yourself.
Building their first AWS project by vibe coding, the writer designed the architecture and let Codex write the code. A flawed first design took a day, against an estimated three or four; the redesign also took a day and scaled.
Result ChatGPT's thinking mode caught a concurrency bug
A roughly 50-page tax return, run through three coding agents side by side. Codex was accurate first time in 12 minutes, Antigravity missed a value until prompted, and Claude Code stumbled overnight.
Result Filling the forms by hand would have been faster anyway
Four years after shelving a homemade RSS reader and selling off its $500 domain, the writer built it this year with Codex and Claude. They use it daily, along with 30 other strangers who found it public.
Result 30 strangers now use it too
Nap-trapped by their baby, they built an MCP bridge so an agent could run their Ableton Live set directly. Codex composed an EDM track from scratch, then they revised vocals, drums and dynamics with follow up prompts.
Result Also had Codex compose a piano duet debating AI's merits
A client wanted a Belmond hotel in Italy, and the advisor had only ever stayed at one Belmond, in Anguilla. Codex, pointed at the client folder, found the notes, checked Belmond's live site and ranked three options.
Result Found the right notes file without being told which one
A master's thesis in a fast-moving field and too many new papers. A weekly Codex automation searches arXiv, triages only new papers and files promising ones into Zotero, with fewer hallucinated citations.
Result Sorts papers into keep, skim, drop or duplicate
Six AI workers, two each on Claude, Codex and Gemini, were set loose to build a SQLite-like engine in Rust. It passed 64 queries of a 1,000+ query suite, with slop like redundant buffer copies and no concurrency.
Result 54.5% of all commits were just lock coordination
"What's for dinner?" with allergies, expiring food and a garden to use up. ChatGPT wrote the build plan and Codex built a local app. Two restarts and about 20 hours later, it's an MVP they use daily.
Result Codex claimed at least 3 of their ideas as its own
Devialet's official app was abandoned. Claude and Codex reverse engineered the amplifiers' private protocol and built an iOS controller in about a day of work. Without AI, they'd never have started.
Result About a day of work, spread over a week
A poorly designed Chinese API that had to be polled again and again. Codex built the integration (with human guidance where "AI was lost"), the load tests failed, and they advised backing out before a big deposit.
Result Advised backing out before a big deposit
A new HDMI-CEC controller didn't work with their projector setup. From their phone, Codex diagnosed the firmware, rewrote it, tested it and flashed the fix over the air.
Result Diagnosed, rewritten and flashed over the air
Buildkite wants GitHub webhooks for every repo, the kind of setup that's tediously annoying by hand. Codex built ephemeral runners on refurbished Mac minis and migrated each repo. Payback: two to three months.
Result Their colleagues are now considering it too
Reverse engineering the firmware meant soldering wires to UART headers in an awkward spot. With Codex it took about 10 hours; without it, they say, it would have taken many more weekends.
Result Works fine now, after awkward UART soldering
Two seven-seat cars for a Maine trip. Codex browsed Avis, Hertz, Costco and Chase, queued the CAPTCHAs, and found a reseller at $600 against $1,100 direct. The writer counts $1,000 saved for both cars.
Digging through SaaS APIs one record at a time is slow for an LLM. So a small company dropped several SaaS tools for self-hosted ones and plain files, and cut licensing fees on the way.
Result Lower licensing fees, faster data access
They already run local reviews with Codex and Claude Code. Assigned as a reviewer in GitHub, Copilot finishes in a couple of minutes and raises valid issues they'd never found.
Result Review done within a couple of minutes
They plugged the board into their Mac and told Codex what it was, how the motor was wired and what it should do. Codex wrote all the software, and it worked exactly as wanted.
Result Codex wrote all of the software
No OpenClaw needed: Codex on a ChatGPT Plus plan, kept alive by systemd, with a UNIX account, a mailbox and a diary for memory. It does all kinds of company tasks, and a second bot may follow.
Result No memory problems so far
Code from one AI is hard to trust blind. This developer runs a Docker harness where Claude writes and Codex reviews, works in it most of the day, and open-sourced it as moarcode.
Result Open-sourced as moarcode
Optimizing a Taichi kernel had each agent looping on its own. Putting Claude Code and Codex in one discussion to combine their ideas cracked it, saving several hours.