Scott Purdy

@KentonVarda Completely agree. Very hard to find scroll wheels with a good click and the sensitivity buttons are useless. The best I've found is the Glorious Model D (can't speak to the D2 or D3).

Logitech in the last few years changed their mouse wheel such that it's difficult to click the middle button without accidentally scrolling, which makes it basically unusable. And every gamer mouse these days seems to have at least 5 different buttons to control sensitivity, which I never use, so it's just obnoxious. I just want five working buttons and a scroll wheel.

1999 Microsoft IntelliMouse Explorer, still the greatest mouse ever made.

@simonlast Do you have any tips on how to set up the reviewer? And how do you split review and testing?

6/ Adversarial review is what makes long, unattended runs possible. Before any task is checked off, a fresh read-only sub-agent reviews the diff against the todo + plan and return gaps. (This is often too powerful and you need to dial back from over-engineering).

4/ Drive it with a persistent task list. Your job is to add vetted tasks faster than it can complete them. It's like shoveling coal into a steam engine. Each task should be: what to do, how to verify, proof notes once completed. The bar is: if satisfied as written, you'd trust the result. Queue as much work as you can before you stop working for the night, especially on Fridays.

XO lives at exoh.ai. Let me know if you’re interested in trying it, are also working in declarative agent frameworks, or otherwise want to connect. I’m planning to write more detailed deep dives when I can pull myself away from building long enough to do so! (7/7)

Context windows are limited so scaling workflows requires chaining agents with constrained context. It requires specialization in multi-agent workflows, repeatable routines, and essentially building an OS for agents. (6/7)

The bottleneck in human-agent workflows isn't just model capability; it’s human attention. I’ve been running an experiment called XO (Executive Officer) to explore ways to scale agent usefulness per second of user attention. 👇 (1/7)

@floydophone Is that an upgrade?

new claude code loading screen leaked

Tomorrow, we launch. At sunset tonight, Artemis II waits on the pad, ready to carry astronauts potentially farther than any humans have traveled in more than half a century. The next era of exploration begins.

Voter ID laws are critically important but it must also be free, easy, and fast for citizens to get the necessary documentation. Right now many places are in the exact opposite situation.

@floydophone A bit weird when you consider that one of the first things that exist for any language is a parser/compiler... Are they just not written in a way that you can build tools on top of them?

seems like every time a really good language parser becomes available the ecosystem rapidly modernizes saw it with esprima/babel in js world seeing it now with sqlglot in data world

@matt_bc @beacon_a_i You said no photos.

A little late to the flag posting party but here is @beacon_a_i’s I once worked for a tech company in SF that had all sorts of flags but HR told the vets group that the US flag could not be flown because it would be offensive to employees. 🇺🇸🫡

Never let your conviction exceed your understanding.

Great idea. This needs to happen!

So much better than Google/Alexa/Siri where you end up memorizing the exact sentence structure needed and one wrong word completely breaks it.

America is the greatest country in the world 🇺🇸

@paulg @lexfridman Nearly 7% of humans are alive right now so I don't think it's much more than 90%. May even be less.

@lexfridman Way, way more than 90%.

Most people who have ever lived are forgotten. Most of us alive today will be forgotten. The lasting impact we have is through our connection to other human beings: through friendship, parenting, mentorship, friendly competition, collaboration, and love.

Docker Desktop on Windows with WSL 2 is an amazing web dev setup.