125
Matt Galligan

@mg #125

Co-founder XMTP, building web3's secure messaging layer ➡️ https://xmtp.org
44401 Follower 256 Following
I asked o3 to "roast me" and was properly charred to a crisp. My. God.

🔥Fire up ChatGPT, select o3, and type "roast me." But brace yourself…
https://imagedelivery.net/BXluQx4ige9GuW0Ia56BHw/661b34c0-bcf2-4592-190c-882fec49e100/original
Running a poll for people that use ai agent software dev tools. If that's you, please toss in your answer!
Loading...
Poll: For those that you who use AI agent software dev tools, how many different ones do you use in a week's time?

Tools like Cursor, Windsurf, Claude Code, Cline, Roo Code, GitHub Copilot, VS Code Agent, Aider, etc.

Reply with which one(s) you use!
Anyone else use n8n for automations?

I couldn't seem to find a node that connected with Warpcast/Farcaster for posting, feed/profile fetching, etc. Going to try and put one together, using the @neynar APIs.

Hoping to set it loose next week. 🤞🏼
Cursor is a wrapper for ChatGPT. Also:

🌯 Burritos are wrappers for beans, cheese, and meat
💿 Playlists are wrappers for songs
📰 Publications are wrappers for writings

The parts can be enjoyed on their own. But wrappers can also be enjoyed as more than just a sum of parts.
Pairing with AI agents to code feels like immersion language learning to me.

I know the output I'm going for when prompting so when the result arrives I can look through it, and try to understand the patterns & techniques that were used.

It's gotten me so much further than any code tutorial ever has.
Loading...
Building w/ AI agents has been transformative for me.

Feeling like I can finally unlock ideas & explore things without being held back by my coding…

But actually learning things has been my favorite part.

So while you vibe away, don't miss out on the teaching moments. 🧠🧐📚
VS Code walked so that [???] could run.

What's the next evolution of an IDE going to look like that's not a fork?



https://warpcast.com/mg/0xa0f5f058
Loading...
I wonder who the first team will be to break free from the VS Code dependency, and deliver a fresh take on what an Agent-forward IDE could be… 🤔

VS Code is great, but is kind of bloated, and wasn't really designed for the workflows that could make agent-driven development shine.
Thought it might be fun to start open-sourcing and posting the AI prompts that I'm making and using daily. Just put up my first one, which I've had a lot of fun and success with—essentially an ideation partner.

Check it out and let me know what you think: https://warpcast.com/mg/0x3f73930b
Loading...
Though it might be fun start open-sourcing AI prompts, instructions, custom modes, etc. that I make and use regularly.

Some of these have been a game-changer for me, so hoping that they might be helpful to others! Try them out and let me know what you think…

👇🏻 Here we go 👇🏻
Me waiting for Muppets to overtake Ghibli
https://imagedelivery.net/BXluQx4ige9GuW0Ia56BHw/5f612c0b-e6e8-4809-6f2e-26fceb076e00/original
The new GPT-4o image generation model is just plain fun.
https://imagedelivery.net/BXluQx4ige9GuW0Ia56BHw/9d862478-b431-4bd7-dc97-4cc026903b00/original
Made something, would love feedback!

It's a markup spec / ai prompting technique I've been using with some early success. I designed it to be straightforward to write, make use of variables, and produce consistent results.

Try it and let me know what you think!

https://warpcast.com/mg/0x20b9cae5
Loading...
/aichannel
What do you prefer to use in your LLM prompts? Mostly curious for those hoping for a specific output.

https://frame.weponder.io/api/polls/30211
Anyone out there that's got some expertise in doing AI prompt evaluation? 🤞🏼

I've been experimenting with a new prompting technique that's showing promising results, but really want to put its through its paces.
Just got through Brussels customs. Wow that was a real dumpster fire. 🛃

…but then a customs agent pulled my wife, 4mo daughter, and I out of line, saying he didn’t want the baby to wait.

Dude literally had left his customs window after seeing us, just because. 🥹
/EthCC
While it’s by no means a scientific measure, lately I’ve found that Claude 3.5 Sonnet has been producing much, much better output than ChatGPT 4o, which is really saying something.

Anyone else had similar experience? Different? What’s been better for you?
Page Properties are really nice since becoming first-class. One thing I’m still hoping for though is a List property that is filtered based on a defined tag.

That way “People” property could suggest pages tagged with #person…

Or a “Status” property could confine suggestions to #status/…

cc @kepano
Travel is booked, and I’ll be heading to Brussels for EthCC! Looking forward to jamming with y’all on group chats, /xmtp, and making web3 work for everyone.

Any get-togethers or events that I can’t miss?
/EthCC
A MASSIVE upgrade to /xmtp just arrived on the production network: Group chat is now live!

It’ll take a bit for clients to be updated to use groups, but devs can jump in today.

That’s not all though—it’s also a major security upgrade with improved privacy!

More here: https://warpcast.com/xmtp/0x0187bc25
Loading...