Daniel Petro

@marckohlbrugge 😂 that was super entertaining to watch. It felt like it was more of a distraction though and dictated what you ended up talking about because you kept correcting it. Super cool either way! Let me know if I can get early access to leg list 😁

Yesterday some speakers at Ruby on Waves mentioned they vibe-coded their slide deck, which got me thinking: What if you have an LLM vibe-code your slide check DURING your presentation? Based on what you say. So naturally, I built it…

@ivanburazin Are you using Claude Code as the harness with both? If not, could it have something to do with the CC system prompt?

@moboudra haha will do!

@DanielPetroAI looks like you were destined to try it, let me know how it goes!

in Paseo, you can just merge worktrees back to main when you're ready. no github. no remote required.

@srinitude @nicopreme thanks for the laugh!

@nicopreme Shell yea

pi-interactive-shell has monitor mode too now. Tell Pi "watch my tests" or "let me know when a PDF shows up." It runs headless and only pings you when there's a match. pi install npm:pi-interactive-shell github.com/nicobailon/pi-…

@llmdevguy I'm testing it with Gemma 4 26b a4b and trying to narrow the use case down to specific subagents in order to squeeze the most juice out of it. No clear direction just yet but I think it's definitely worth exploring. Would love to learn from you or others' experience!

🤔PI coding agent + local model like Qwen 27b sounds like a good idea. PI is very token and tool calling efficient. So it can squeeze better results from local models. At least I think so.. Anyone tested this?

@badlogicgames @dillon_mulroy Congrats! The article really resonates with me and I have zero doubts Pi will get even better under Earendil. Thank you and all the contributors for Pi!

People of pi. BIG NEWS. I've sold out. Let me know how you feel about this in the comments below. mariozechner.at/posts/2026-04-…

@daosterman @llmdevguy Me too! I haven't had a chance to fully configure my setup and push it yet but I'll definitely share my experience when I do. Got a little sidetracked with testing Gemma4 31b locally with Pi lol

@DanielPetroAI @llmdevguy I really enjoy using codex. I also agree with your sentiment about the subsidized subs. I need to explore some other options as well. How are you liking the Go plan on OpenCode?

The honeymoon phase is over

@bohdanpodvirnyi Super cool and great timing!

people of pi (and other agentic beliefs)! may I present you a tool that mirrors agent sessions between harnesses so you can easily resume any conversation with any agent pi <-> claude code <-> codex github.com/bohdanpodvirny…

@merlindotcom_ @Paperclip_AI Followed just in case you do, appreciate the reply!

@DanielPetroAI @Paperclip_AI Custom fork. Can’t merge back. Departed too far… Might release under a new name but unlikely to be office tools themed haha

pov: you added all the features you wanted to @Paperclip_AI because you couldnt wait. now you run your whole marketing team, dev team and EA all from one place. you also use it to improve itself. never had so much fun

@mattyp Incredible work and thank you so much for all that you've done! I hope you enjoy your time off and can't wait to see what's next for you

Today is my last day at Replit A friend of mine once said "you're never on the same team twice." I'm grateful for every single version of the Replit team I worked with I'm grateful for our users, some of the brightest and most determined out there I'm grateful for the progress we've made toward our mission to help everyone express their ideas I'm grateful for @amasad & @HayaOdeh, who gave me a platform to do my best work, @davidhoang, who took a chance on me two years ago, and @pirroh, my manager I'm grateful for countless current & former employees who I call my friends Everything I did started with our amazing team & product. None of it was possible without them I'm taking some time off to touch grass, then it's back to building Thank you for following along - I can't wait to keep pushing Until next time ✌️

@llmdevguy I thought it's against TOS to use it with a CC sub? Either way, would love to hear how it works out for you!

@DanielPetroAI I will test pi this week. It may be good with CC to save tokens.

Anyone using Pi or OpenCode over Claude Code/Codex? why?

@llmdevguy I will as soon as I find the right harness for it! But eventually, I don't want to rely on any of these heavily subsidized subscriptions because then I'll be in the same spot again. I'm trying OpenCode's Go plan in parallel

@DanielPetroAI Switch to codex. ;)

@charlieholtz Woohoo! Any chance y'all will support opencode or Pi?

we are now boarding the Conductor express

@mattpocockuk I haven't tried it out yet but I love the name!

I couldn't find a great, simple framework for parallelising sandboxed coding agents. So, I built one. It's called sandcastle. This `run()` function: - Creates a worktree - Runs a sandboxed coding agent (on Docker) Super powerful, super-simple, 100% local

Yeah it's great, but definitely needs more encouragement and isn't a natural go-getter. Also, I know this isn't true, but it feels like even the Plus plan is getting me more value than the MAX 5x plan lol.. I downgraded from 20x last month because quality wasn't great. Bonus, you can use it with other harnesses!

Ok enough!! time to cancel my claude subscription is codex any good?

@karpathy Did you try your llm-council for this? I don't use yours but I've found multi-round council debates to be pretty effective!

- Drafted a blog post - Used an LLM to meticulously improve the argument over 4 hours. - Wow, feeling great, it’s so convincing! - Fun idea let’s ask it to argue the opposite. - LLM demolishes the entire argument and convinces me that the opposite is in fact true. - lol The LLMs may elicit an opinion when asked but are extremely competent in arguing almost any direction. This is actually super useful as a tool for forming your own opinions, just make sure to ask different directions and be careful with the sycophancy.