Hook

Their other posts in the index, biggest breakout first.
Okay. The moment we've all been waiting for. Fable 5 is back. The only 50% of your normal usage limits and most of you are about to use it wrong. Anthropic wrote the manual last time around and almost nobody read it. Let's cover the five ways to use it right and two ways not to get caught. And hey, if you're new here, I'm Kevin. I break down what's happening in AI and how to build with it. If that's your thing, hit follow. Okay, so we're back. Fable 5 is here. And the one thing you actually need to do is read the actual prompting guide. Yes, read it. That's the part everyone skipped. The rest that comes straight from that guide. Your usage limits will thank you later. Number two, prompt the goal, not the steps. What does done look like? Why does it matter? Where are the limits? And how does it check its own work? Treat Fable fine and give it some space to breathe. Number three, gut your Claude.md file. Those always read the whole file first rules you wrote for the old models like Opus and Sonnet make Fable act dumb. Babysitting. Delete the babysitting. And this is a good opportunity to review your Claude.md files anyway, they're to make sure they're not over 200 lines. Number four, know your effort dials. Default high, extra high for hard stuff, skip max, it's not worth the cost and then overthinks. Ultra one, you can code extra effort and run the whole agents, a whole team of agents to work on the job. Number five, give it a big ugly job. Refactoring, architecture calls, a bunch of nasty bugs. The small stuff and to Opus and Sonnet, you're overpaying. Now, don't get caught with your pants down. It's $10 bucks in, $50 bucks out per million tokens. Free 50% of your weekly limit until July 7th, then it's metered from then on. One guy, I saw an X, torched 150k tokens on a single question. That's expensive. It's also more touching than before. At least 5% of chats will quietly reroute the 4.8. And right now it looks like, and biotech and cyber filters refuse normal stuff like reading your blood work or checking the security of a code base. So don't ask it to do any of that work or print its own reasoning. It will likely refuse and kick you back anyway. Let me know in the comments the first thing you're going to sick.