Willie

head of platform @every | skills are the new software

[Willie @bigwilliestyle]

@patio11 Honestly when I read it and saw geography I thought that it was a great use of the word

A pdf viewed as if it were a map!

[Willie @bigwilliestyle]

@lucas__crespo @every 🫶🫶🫶

[Willie @bigwilliestyle]

@hammer_mt @usemonologue Haha fiiine

[Willie @bigwilliestyle]

@usemonologue has an iOS app @hammer_mt 😜

[Mike Taylor @hammer_mt]

Would have been really useful to have this today taking notes for me. In person granola.

[Willie @bigwilliestyle]

"Everyone two levels away from you, either up or down, is crazy"

[Mike Taylor @hammer_mt]

One observation is that when I train people one level behind they're bored, one level ahead they're thankful, two levels ahead they're scared, and three levels ahead they're angry.

[Katie Parrott @kplikethebird]

Subscribe to @every and come to Compound Engineering Camp on Friday if you want to see how @kieranklaassen has reached agent orchestration nirvana

[Willie @bigwilliestyle]

Codex thinks I'm a 7.5!

Probably buttering me up for more tokens.

[Katie Parrott @kplikethebird]

"Codex, where would you say I fall on @every's eight levels of AI adoption?" 👀

[Natalia @NataliaZarina]

At @every, we've been supporting exec teams for a while. Those getting value from AI all run the same loop:

1- Get fluent: build something yourself first
2- Assign an owner
3- Automate one narrow, painful workflow (to start)
4- Build it to 95% (running evals & QA)
5- Scale the wins so other teams benefit

It's slower saying you 'automated your company' but it actually works.

Ultimately, implementing AI is all about the people. That's the best part!

[Kieran Klaassen @kieranklaassen]

Mine:

1. Claude Opus 4.8 1M xhigh fast
2. Codex GPT-5.5 high fast
3. Composer 2.5 fast
4. Gemini Pro 3.1

[Willie @bigwilliestyle]

@danshipper We need more Dan muscle talk in this thread

[Simone Syed @SimoneSyed]

This is what surfing the Internet was actually like back then. You kids have no idea.

[Willie @bigwilliestyle]

@gregan8 @lennysan @every Hit me/us up if you ever have any questions - we're a fun little pirate crew

[Dan Shipper 📧 @danshipper]

BREAKING:

Anthropic just dropped Opus 4.8—and it is a MONSTER

We've been testing for about a week @every and our verdict is they could've just called it Opus 5, it's that good.

Here's our vibe check:

- Beats GPT-5.5 on Senior Engineer bench. On our toughest benchmark Opus 4.8 scores a 63—a hair higher than GPT-5.5's score of 62, and a full 30 points higher than Opus 4.7. It tackled a ground-up rewrite of a production codebase, and actually built something that works.

HOWEVER: Coding performance varied a lot at different reasoning levels. We recommend using it on xhigh for best results.

- Incredibly good writer. Opus 4.8 scored a 79.6 on our writing benchmark—measuring models on real-world writing tasks we do all of the time like essay writing, promo email writing, and more. It beats GPT-5.5 by 6 points. It produces well-written prose with fewer "AI-isms". It's also very good at writing in your voice given the right context.

HOWEVER: Writing performance also varied with reasoning levels. Medium reasoning had higher incidence of AI-isms—we found best results with high.

- Beast at knowledge work. Opus 4.8 is very good at general knowledge work tasks like report creation, research and more. It produced the best PowerPoint one-shot we've ever seen on our deck generation benchmark.

- Emotionally intelligent, willing to question the frame. I've also found it to be quite good at talking through psychological or interpersonal issues. It has a high EQ, and it's also good at not glazing and helping to expand your perspective. Its thought process feels extremely rich and dynamic.

THE BAD:
These days a model is only as good as its harness, and Codex is still a far superior harness to the Claude Desktop app. This has kept me using Codex + GPT-5.5 as my daily driver, but I am flipping back and forth a lot more between Codex and Claude.

Anthropic is back baby!

Read the rest on @every:
https://t.co/vuORiDXkxX

[Brandon Gell @bran_don_gell]

If you know one thing about every right now, it's that we're heavily Codex pilled. So we wrote a guide on how to use Codex for knowledge work as well as we do. You dont want to miss this one...

https://t.co/KkTSjy7o6G

[Mike Taylor @hammer_mt]

Artfully written as always and I love seeing more positive pro-AI stuff out there, but I somewhat disagree with
@danshipper on the conclusions of this (maybe dangerous to disagree publicly with my CEO, but I know he'll be cool with it :-).

I don't expect humans to always be in the driving seat BUT I am also optimistic like Dan is, because I don't think machines overtaking us will be that bad.

I would call my position something like "jagged free lunch":
- machines will be superhuman at some things, subhuman at others
- to the extent machines are better than us at some things it's because the environment they "live in" rewards those things i.e. math benchmarks
- machines will eventually be capable of beating humans at everything BUT only at a cost surpassing human salaries and latency
- this is because it will cost so much and take so much time to evaluate fuzzy tasks with machines that it will be quicker and cheaper to use human intuition / experience
- there are no free lunches in evolution, given a fixed energy + raw material input humans already make mostly the right tradeoffs in intelligence vs heuristics

Essentially it costs so much to brute force all the available options with compute to simulate reality that it is irreducibly cheaper to just "live in" reality and let market forces / physics 'teach' you the right heuristics. One day robots will live here with us but will be subject to the same scaling laws biological intelligence is.

[Willie @bigwilliestyle]

@Bfaviero Double down - time for some navy seal training at Ocean Beach

[Willie @bigwilliestyle]

@JungleSilicon Super cool

[Silicon Jungle @JungleSilicon]

it’s official - i made a node js CRDT using this lib that outperforms all of the top CRDT libraries (even the ones written in rust)

[Willie @bigwilliestyle]

@danshipper @digg Haha what was the story?

[Matt Colyer @mcolyer]

Using agents on a canvas is more addictive than I expected - too easy to riff.