And yes, you can download both Shelf and Ledge. Tell me which one you like better and why in the comments!
Better computer use and longer-running agents make more possibilities worth exploring. What to reconsider, and how to give Astra the work.
What changed this week, with GPT-6 Astra and Fable 5.1, is how little you have to say to start one, how long it runs without you, and what your job is when it comes back.
Claude Fable 5.1 is a knowledge work monster!
My $600-plus AI setup, what I keep outside every provider, and the test each expensive plan has to pass.
What 1,200 experimental agents, a $21 million startup, and one missing advertising account reveal about the hardest part of putting agents to work.
The dangerous version of brain rot looks like productivity: more shipped, faster, and fewer decisions that are actually yours.
The work agents create has no name, no owner, and no line on any dashboard — and learning to manage this work is starting to decide who gets promoted.
The forward-deployed engineer is what every AI lab is hiring while promising autonomy. Your industry background is the part that qualifies you.
Keep the coding harness you already built while GLM-5.3 moves expensive API work to a much cheaper model in minutes.
Build it! The complete guide for non-developers: pick the right shape, make the few decisions that matter, and turn one stubborn wish into working personal software.
Your team isn’t asking whether AI tools work. They’re asking what happens to them if it does.
An agent attached the wrong file to my email and reported success. I almost hit send. Here are the three checks I run now, and the question I answer before all of them.
The people who take the time to read with care are paying for the people who don’t. Here’s the skill I built to keep your own work out of that pile — and why no skill can do the part that matters.
Aschenbrenner read AI right and still had to sell. Apple has looked slow for two years and it doesn't matter. The forecast was never the problem.
A five-level map for figuring out whether you have an idea, a business, or an advantage that can survive the next model launch.
You have a file you would never paste into a chatbot. We all do: a contract, a board deck, a client record. AI could obviously help, but you also know you can’t send it to a model provider, so the work stays manual, or it doesn’t happen at all.
You can use Kimi K3 on Moonshot AI’s website today, much as you can use ChatGPT or Claude. On July 27, the Chinese company says it will take a second step: publish the model files that let other companies run and adapt K3 on their own systems.
The two ways agents fail you. Understanding theater, where a good conversation convinces you the work was understood, and completion theater, where a finished run feels far more done than it is.
Claude Opus 4.8 is excellent. The harder question is where it should replace your current workflow, where it should be a specialist, and where turning the reasoning dial up can make the work worse.
