Compound engineering and the Agentic loop
dimillian.medium.com/the-agentic-...
Compound engineering and the Agentic loop
dimillian.medium.com/the-agentic-...
My thoughts on the recent fast-paced evolution of our AI workflow: The Tipping Point
dimillian.medium.com/the-tipping-...
Just Codex
The future is literally at our fingertips. I believe we're almost there with the model if we build enough tooling to become autonomous around it. And if we're patient enough, we'll get the model that will do all of it itself anyway.
But this will be solved soon. If I had enough time, I would code an autonomous Codex GitHub agent that basically takes issues one by one, implements them, tests them, or closes them.
What's the point anymore? Codex literally reviewed 500+ open issues, grouped them, and sorted them for me in less than 3 minutes. Everything became easy. It's just a matter of will, drive, and ideas.
The only bottleneck is me having to manage all that
About time!
Adjusted my setup a bit today
I keep seeing people say they'll get rusty if they let the AI write everything. Well, the truth is that you can't compete anyway. It's time to sharpen your agentic skills rather than refine your decades-old programming skills.
Itβs a bit like that the point? You can still read the code and steer it to what you want the code to be. But writing code in itself is already almost all gone
One thing that didn't change with agentic programming is that good software is still obsessing over every little detail. Both technical and UI based. That's still the fun part for me.
And I'll continue to maintain mine; both can cohabit. I don't see a problem with it.
And the Codex team has been amazing with all the feedback they've gotten on their own app, and it's already steadily improving!
All in all, it's a very solid first release! I can't recommend it enough; this is truly the flow of the future, and I'll only improve from there.
It has many different features from Codex Monitor. I see my own app as a simpler/leaner version of the Codex official app.
And it also has all the features you expect from this kind of new, agentic-driven environment. A sidebar with the git diff changes, a terminal, a way to quickly add and sort workspaces, easy access to all Codex /commands, plan and code mode, etc... And multiple run commands!
About worktree: the Codex app has a very cool feature. Once you have changes on a worktree, you can sync the change to your local repo, it'll create a branch for you and keep it in sync between your worktree and repo.
This is amazing and makes Worktree much better to work with!
Another feature they have is automation! You can create an agent cron job that runs on a specific day at a specific time. In my case, I have a daily review of my open PR and issues, but what you can do is pretty limitless!
Another awesome feature is that you can easily start new agents locally, in a worktree, or as a cloud agent. To me, that's a super powerful feature! Your worktree and cloud agents are combined in the sidebar. And you can easily apply cloud/worktree changes on your local repo.
You can also manage and edit your skills and install new ones from an ever-growing marketplace of skills! Also within the new Codex app!
Can't wait to see where this will go. Skills have been a bit messy, with a ton of libraries and repos all over the place.
You can see all your agents' threads per workspace in the sidebar and quickly switch between them.
It has most of what I ever wanted from those new kinds of agentic/orchestrator IDEs.
And no, no code editor, we all agree on that!
First: you can download it here: chatgpt.com/codex
Now, about some of the featuresπ
I've been testing the Codex official macOS app for the last couple of weeks. Yes, funny, I know! But I didn't know they were making their own until a week after I started and released Codex Monitor!
It's OpenAI's take on the IDE of the future, and it's pretty good!
π§΅for more
That was a solid (tokens) week
Someone is playing with the idea on the repo
A sincere salute to the amazing folks who generously imparted their invaluable insights to us: Shili, Snow, Mohammad Azam, @marcprux.bsky.social , Khoa Pham, Zhong Cheng Li, @joanniso.bsky.social, @pepicrft.bsky.social, @dimillian.app and jaywcjlove
For those who are not terminally ill, I've created a macOS Codex Application that can manage any number of agents across multiple projects with a sleek UI.
And of course it's fully open source!
github.com/Dimillian/Co...
Itβs a small trick I used for the conference. Itβs not Claude but cursor auto completion model which inferred it. Because I created and deleted the project multiple times in the same folder cursor had cached previous attempts.
Ice Cubes is one of the best apps to experience Mastodon β free, open-source and constantly improving. But how to deal with 500+ tickets as a single maintainer? π±
Read how @dimillian.app prioritizes features with help of analytics:
telemetrydeck.com/casestudies/...
#DataDriven #AppAnalytics
I'm making an open source Diablo like game using Love2D, feel free to follow along! github.com/Dimillian/Di...
If I do a PR to fix it youβll do an iOS 26 version?