LiBuild

@awsdevelopers The real party will be when AWS turns 21 😄

when I realize I am turning 20 this week

@jessfraz Your CAD tools must be doing some serious magic, when I asked ChatGPT to estimate the dimensions of my parsons table, it couldn’t do that

gpt-5.4 + our cad code language (KCL)

@ShrekOverflow Want the TestFlight?

@libuild 👀 what’s this?

That feeling when you wrap up the day and start seeing the light

@ShrekOverflow +1 😂

The LLMs’ rabid obsession with successfully completing the task is sometimes a little much

@elonmusk The model alone can’t make security decisions, adding external safety nets can make it safer: x.com/libuild/status…

People giving OpenClaw root access to their entire life

@ShrekOverflow @jessfraz This is absurd 😂

@jessfraz Definitely figma side

me: i hate this figma make: what would you like me to do to that element? change it... me: nuke it BIG RED ERROR you are not in line with our guidelines me: delete it GODDAMN WHEN DID YOU ALL GO SO FUCKING SOFT

The policy engine evaluates the agent's actions in realtime by understanding the intent of the action. As a result, you can simply write "don't take any destructive actions" and the policy engine can infer what this means and apply it globally to all tools.

Prompts are in English, so policies should be in English too. But that doesn't mean put your policies in your prompt, here's a better way

@jordienr I want to be guided by a wise old crone

@libuild A what

QUESTLOG IS IN THE APP STORE FINALLY

20-50% of people would just pick “Other” here

What in the security antipattern hell is this

@ShrekOverflow Pretty cool, where is the code?

I needed Google Docs like collab mode for my MDX blog, so I just built it in 2 hours with Cursor. All code is in MDX, Comments and Suggestions, Resolutions all get stored in git. A small MCP server is all it took.

@positiveblue2 @cursor_ai @embirico Is your work more brainstorming or more editing? I find that the GPT style models tend to be great at thinking v/s Claude is almost too eager to just do.

Total spend for the period (Jan 12 - Jan 21) in @cursor_ai 🥇GPT Extra high: $36.42 🥈Codex Extra high: $18.49 🥉Opus: $12.77 Total: $67.68 ($6.77/day avg) GPT extra high is doing the heavy lifting for me (more than Codex @embirico) mainly in Ask mode

@ibuildthecloud At least we did Consent Engineering before Harness Engineering

Harness Engineering!? Can we stop just throwing a new noun in front of engineering each week.

@yenkel Good catch! Apparently it’s intentional: …symbolically it’s: overworked but game, slightly caffeinated, and still trying to be helpful. If I were being extra honest: The smile + headset combo is “this is a lot, but we’re doing it anyway.”

@libuild ha. curious about the headset. asked it to explain?

tamos safe

TIL that people in 8,000 BC had better auth technology than a lot of software today: en.wikipedia.org/wiki/Bulla_(se…