Hey fellow friends! Excited to spend the week building with Fable 5. Spent the weekend euro-maxxing, did I miss anything?

Would love to see this benchmarked on @cursor_ai CursorBench! /cc @ericzakariasson @leerob

„We can keep being the continent that writes the most thoughtful rules about a technology it does not possess. Or we can become a place where the frontier is built, and not just licensed.“

first thing I do when Dario publishes a new essay: Send to Kindle so I can read it there!

Claude Fable 5 is here. Anthropic’s new model is extremely capable, extremely expensive, and not quite like the Claude releases we’ve seen before. I made a video breaking down what it is, why it matters, and how to think about using it:

good way to think about when to use a bigger model in general, but especially true for Fable 👇

You gotta think in terms of systems. You use the results and review to add in quality controls and improve the odds of success over time while scaling. You add validation testing and additional prompting in the process. Add a fine tune loop if you want once you get it how you want and learn to teach it to one shot it once you gather enough data. Scale might initially produce more slop but then you build in additional polishing loops. Putting Cursors thermonuclear code review refactor and also using it in a debug loop like @RBouschery did github.com/rbouschery/agent-… plus adding autoresearch or /goals to loop it or adding your own algorithmic process help. Add “field testing” where you prompt agent to actually generate tasks to prove it works in the real world and then add “human in the loop field testing” where you validate the test works. You can probably replace that too once you train RL to predict human review near indistinguishable from human review itself. You gotta learn to replace yourself and keep working on higher abstraction layer and then figure out how to replace that too.

Both of these statements are true: AI tools democratize professional work. But seniority still matters. A beginner can now create things that looked impossible a few years ago. They can design an interface, generate a professional slide deck, write functional code, or edit a video. Give the same tools to a senior engineer, designer, analyst, or strategist, and the output will still be much better. Not because they can click different buttons. Not because they have access to a secret model. But because they know what good looks like, and they have the language to describe it. A lot of professional skill is vocabulary. In design, “make this bigger” is vague. “Increase the padding, tighten the hierarchy, and reduce visual noise” gives the model a much clearer target. In creating slides, “create a chart” is vague. “Use a dumbbell chart to show the before/after delta across categories” gets you much closer. The same is true in code. The same is true in design. The same is true in almost any form of knowledge work. Language matters because language carries judgment. It encodes distinctions. It gives shape to vague taste. And as more of our work becomes instruction-giving to agents, this matters even more: ​The quality of your output increasingly depends on the quality of your instructions. This is why I think agent skills are so interesting. At the most basic level, they let you borrow the language of experts. You can take a skill written by someone who understands design, code, research, writing, or slides, plug it into your workflow, and immediately give the agent better instructions than you would have written from scratch. That alone is valuable. It makes your work with agents more effective. But the bigger opportunity lies in taking one more step: If you do not just use the skill passively, but read it, question it, rewrite it, and adapt it to your own work, you start upgrading your own language. You begin to notice the distinctions experts make. The defaults they encode. The constraints they care about. The words they use to turn a vague request into a precise instruction. Over time, this changes how you prompt. But it also changes how you think. You are not just copying someone else’s expertise anymore. You are absorbing it, reshaping it, and combining it with your own point of view. ​ So don't sleep on agent skills and don't just use them passively.

interesting pricing observation: for a long time, the yearly plan was chosen ca. 80/20 vs monthly option on @readbetterio in recent months this flipped - now most people choose monthly. i too commit to software much shorter because my own usage patterns change so fast. anyone else noticing this?

I love X but @nikitabier wtf is this - please fix ads next - I have to report several ads a day that are the worst spam

look ma, I’m a billionaire! /cc @cursor_ai

cool! Joined 821 days ago - wow didn’t realize it’s been that long! cursor.com/@rbo

I love how the Cursor tab button seems so much like a relic of the past - almost like showing a kid today a landline phone. Remember when we were AI-coding by accepting suggestions by pressing Tab? Ancient times ca. 15 months ago

AI tools can tempt you to work at 100x speed. ​ And honestly, I’ve been there. ​ Spinning up ten agents in parallel. Jumping from tab to Codex to terminal to Cursor like a madman. Moving fast. Getting a lot done. Seemingly. ​ Then reality kicks in. ​ Your mind can only go so fast. ​ And if your goal is not just to generate output, but to actually ship something good, you cannot simply hit send on everything the AI gives you. ​ You have to review it. Understand it. Challenge it. Spot the holes. Sometimes take two steps back. ​ Very quickly, all that speed can leave you with surprisingly little to show for it. ​ I think there is a better way. ​ A few days ago, I was out for a run. For the first time, a voice inside my head said: ​ Go slow. ​ So I slowed down. ​ I glanced at my Apple Watch and saw my pace drop below a “magic” 6 min/km mark. At first, it felt wrong. Like I was underperforming. ​ But then something shifted. ​ I reached a pace where I felt like I could keep going forever. ​ I had read many times that slow running builds the capacity to go far. But my ego always got in the way. I kept chasing some arbitrary “fast” pace, even when it was not helping me become a better runner. ​ And because running always seems to unlock better thinking for me, the analogy suddenly clicked. ​ This is exactly how I should work with AI. ​ Why try to go as fast as possible when the goal is to go far? ​ So I started slowing down in my agentic coding work. ​ One task at a time. One thread of thought at a time. Actually reading what the agent produced. Challenging its assumptions. Staying close enough to the codebase that I could still understand what was changing. ​ And suddenly, I felt like I was achieving more. ​ The code was better. My reviews were faster. I had fewer messy detours. I did not lose the plot halfway through the work. ​ Going slower helped me go further. ​ Maybe that is the more useful version of the “100x developer” idea. ​ Not 100x speed. ​ Maybe it is 10x speed with 10x quality. ​ That still gets you to 100x.

You can work 5 days a week and succeed as a startup. Mercury has done that from day 0 and we are valued @ $5.2bn 7 years after launch. I have been an entrepreneur for 20 years and raised 3 kids while doing it. The point of success is to have a great life not just a startup 😊

a) it‘s not that proprietary as a lot of it can/could still be scraped b) Grok is one of the top models out there and it took him many years less than competition so maybe this proves the opposite?

need some sci-fi book recommendations please - what was the last banger you read?

lmfao these kids really think GMV = revenue

remember when people said SF was dead? looking back it’s incredible how much of a fever dream covid era was