Alexander Embiricos
@embirico
Codex @OpenAI
Couldn't be more excited to share my recent focus: Codex, a software engineering agent that answers codebase questions and drafts pull requests—working on its own computer. This video from @DKundel sums it up nicely. This is just the beginning. Excited to hear what you think!
Here’s how to get started with Codex:
Factory 🤝 Agents
@FactoryAI natively supports AGENTS.md and it's exciting the industry is standardizing best practices for agent-driven development
ChatGPT can now do work for you using its own computer
ChatGPT agent is ready to introduce itself. openai.com/live
This was a useful & interesting read. Anyone else have AGENTS.md setups they'd want to share?
Took way to long to get it in a gist. gist.github.com/brennanmceachr… It's basically a wiki of how to code in various parts of the app. I think have different short specific agents.md files for codex, gemini, vs code, cursor, etc, which direct the agent to read the…
Codex 🤝 Gemini CLI Thanks Kath for going with AGENTS.md from the getgo!
We're working across products to try and make this better for devs. What are your best practices or examples for AGENTS.md? Let us know!
Codex 🤝 Amp. Hats off to @sqs for starting this conversation and being so collaborative. We need more of this in the industry
We will switch Amp to plural. Stay tuned, coming soon, and will have a totally smooth migration path for the filename.
Great news: @OpenAI just got the AGENTS.md domain! Making a microsite on how to use it provide context to coding agents like Codex, Amp & Gemini. The name is intentionally neutral and open. Help me, @sqs, @simpsoka out: Please share your best practices or examples!
in the last 35 days, @OpenAI codex has merged 345,000 PRs on github. 345,000. AI is eating software engineering
Kudos to the Codex team. They fixed the two issues I had: 1. You can enable network access (yay, npm install works) 2. You can push updates to existing PRs Love when teams listen and ship fast.
One of my favorite parts of working with @hansonwng and the Codex research team: Instead of focusing on pure intelligence towards writing functional code, we also pushed on the softer aspects that make code mergable.
The hardest part of building @openai's Codex-1 model wasn't making it the smartest — it was giving it taste. @hansonwng and @embirico dig into what it takes to go beyond crushing coding contests to writing code that engineering teams actually want to ship.
With coding agents, do we increase or decrease the number of software jobs? (📈 IMO) @hansonwng and I had a fun chat with @sonyatweetybird and @laurenmhreeder about the future of Codex. Thanks for having us!
Today on Training Data, @hansonwng & @embirico shared more about OpenAI Codex, how they see the sync and async development experiences merging, and @OpenAI's roadmap for longer-running agents with their own compute environments. Thanks for joining me and @laurenmhreeder!
The Azuki team is always pushing to move faster with AI. One recent upgrade: our marketing team can now ship changes to Azuki.com with minimal eng support using OpenAI codex. They can update copy, swap images, etc... all with AI. The flow: → Marketing prompts codex → Codex…
📣 We're hiring full stack and infra engineers to come work on Codex! If you want to build an agentic software engineer and love both moving incredibly quickly & sweating the details… Drop me a DM! (In-person at @OpenAI offices in SF)