Liam Prevelige

Modeling software architecture @codeviz_ai (yc s24). Views represent my employer

[Liam Prevelige @liam_prev]

@CMonk40079 you’re a bot, but I just wanted a simple human-in-a-box with its own email + phone number

Main things that didn’t work for me: separate history for different input channels, too many decisions about which tools to enable, and I was concerned about malware in skills

[Liam Prevelige @liam_prev]

I gave an agent my license, credit card, root access to AWS, and all the tools I use. It replicated itself in the cloud and I’m running out of things to work on…

[Liam Prevelige @liam_prev]

What I’m most interested in: if you’re about to hire for a role, I’d love to try designing an agent for it first. Not a one-off task, an actual seat you’d otherwise fill with a person. DM me or join our list at https://t.co/G6TUHKGII5

[Liam Prevelige @liam_prev]

Last week I gave it AWS admin access and told it to replicate itself in the cloud. It bought a domain, deployed to Vercel, self-hosted Stalwart, spun up EC2, Stripe, SMS routing. All through texting. I never opened the codebase. The agents now email each other to ship fixes

[Lawrence Chen @lawrencecchen]

Introducing cmux: the open-source terminal built for coding agents.

- Vertical tabs
- Blue rings around panes that need attention
- Built-in browser
- Based on Ghostty

When Claude Code needs you, the pane glows blue and the sidebar tells you why.

No Electron/Tauri. Just Swift/Appkit.

[cat @catheryn_li]

Excited to announce that @simpleailab has raised a $14M seed round led by @firstharmonic.

For the past year, we’ve been building AI voice agents to transform direct-to-consumer sales. We fundamentally believe that voice AI is the future of all B2C calls.

[Liam Prevelige @liam_prev]

@benswerd Prompt https://t.co/9oncKJhHrG

[Liam Prevelige @liam_prev]

The world if we had a computer use model on cerebras

[Liam Prevelige @liam_prev]

It’s really surprising that we don’t have better tools for generating and optimizing prompts. Minor changes in phrasing create huge swings in performance. Even with better alignment/agency this seems inevitable

[Liam Prevelige @liam_prev]

@tenobrus Few coding companies will, and none in long run, because agent quality is converging so quickly

[Liam Prevelige @liam_prev]

@tenobrus It’s still early and they won’t die bc of an inferior coding agent

[Liam Prevelige @liam_prev]

Who’s doing session replay analysis with LLMs? Preferably from yc

[Liam Prevelige @liam_prev]

Needed a banner pic. It was faster to code an interactive 3D canvas than to open Figma

[Liam Prevelige @liam_prev]

You basically train stack graphs

[Liam Prevelige @liam_prev]

This is possible and not *that* hard

[Liam Prevelige @liam_prev]

Is there actually a reason you can’t build good language servers on tree sitter?