Hi 👋, my name is Philippe,
I'm a Software Architect with a specialization in ERP and e-Commerce systems.
Creator of @CodingDodo_ where I write Odoo tutorials with a heavy focus on OWL and the JavaScript part.
I tweet about Odoo, JS, TypeScript, Webdev, and random things.
@devongovett Then we would all lose! Maybe I don’t integrate react aria component manually but I always point my agents to it because I know that’s the top quality one, for what it’s worth I do care even if I a have little voice! don’t give up, you will find pride and fun again
Thank you for taking the time to answer me. I don’t know what to say, now I feel bad talking about it. 😅
I can only imagine how hard it is, with remix history, and I agree, with the current AI era where it seems that we all lost interest in framework « updates », but your blog post was great and it’s starting an important discussion !
But please yes do share more if you can, I am sure a lot of us are still super interested, and stay strong during these changing times
@mattpocockuk I built a little open source, BYOK, based on Pi, cli that generates comprehensive code walkthrough for agent-PR, focusing on understanding intent and critical parts. https://t.co/O2KRyFS4AH if you/anyone want to give it a try 😇
In MVP phase having the agent write ADR often backfires for me. The specs and designs are not fully clear and when you want to change something, agents give me the smug 🧐 "According to ADR-XXX123, we shall not...".
Do I have to enforce a rule that "ADRs should details why architecture was chosen not to establish line in the sand" or something like that ?
"Don't paste the AI, please"
"The world is full of people who don't want to read or think things through. Don't be one of them."
https://t.co/aCuUH0Sgpo
I think we all building the same thing right now but it's super interesting, love the infra as code, agents as files that makes it portable. I am working on something similar open-source, exciting times
Introducing Warp Factories: open, flexible infrastructure for building cloud software factories.
- Configure your factory as code
- Use any model and any harness
- Measure quality with evals and benchmarks on your own data
- Built-in self-improvement and memory
you bring your own model that will create the walkthrough, the mini-harness of balade will load your AGENTS.md/CLAUDE.md and give instructions on exhaustivity of file search so the model will produce something correct.
Then when the model authors the walkthrough it will link to diff hunks, these diff hunks are checked through an internal `check` command so the model cannot link to hallucinated changes or the check command will fail. At that point the model has to retry (with a max of retry based on the budget you choose with --budget flag). This is simplified but overall that's the point
I hope I understood and answer your question, but feel free to ask anything else
working on a little open-source CLI tool called balade to walk you through these giant agent generated PR. Maybe you don't review agents' PR line by line anymore but you should keep a good understanding of what's being done.
It's in alpha, based on @pidotdev, BYOM, uses a Markdoc authoring system that human can also read and fix (if necessary), familiar interface, pin the git head to know when output drifted, and other goodies I am working on, exemple output
I switched to Codex long time ago, I love the app but recently on MacOS it's getting slower and slower. Threads take a long time to open (15-20 seconds) and sometimes freeze the app. I have cron running in multiple projects with heartbeat to check for PR to review etc, and these create extremly long threads that I cannot cleanup, maybe they are making the whole app laggy?
Also the main textarea is getting input lag, the dynamic codeblock by begining to type three backticks is cool but not how I write markdown usually. I generally write ```js or ```python then press Enter, but this doesn't work here you have to do ``` then select the language now that the widget is loaded.
Codex on MacOS is getting slower and slower, some long threads used to take 30s to open, now even medium thread can hang for 10, 15sec. Everything feels laggy, the textarea being the worst offender, latency through inputs and "code fences" bugs made me switch to voice mode 💀
@thsottiaux the latest update to Codex chatbox with code blocks preview by typing ``` is a pain honestly. It's difficult to get out of, Enter goes to the next line, Shift-Enter the same. It's also quite laggy? Can we go back to the old version, we don't need to have this preview in the chatbox please !