Opus 5.5 and Sol 6 both just dropped. Here's a really simple thing to do after you switch. Check how the new model gets around your codebase. Each model finds its way around code differently. Some search first. Some read whole files. Some lean on your instructions file more than others. A repo that worked great with your last model can be wasting tokens on the next one. And that waste adds up. Every extra tool call fills up the agent's context and burns tokens you pay for. Say one common path takes 12 calls when it should take 3. If your team hits it 200 times a day, that's millions of tokens a day spent on searching. You don't have to guess where it's happening. Most coding agents save your past sessions right on your computer. Here's a prompt that's helped me keep my codebase efficient. It works in any coding agent. Copy and paste it inside your repo: --- Find where our past chat sessions for this repo are saved and read through them. If you can't find them, tell me where to look. 1. Find the files and folders agents visit most often. 2. For each one, count how many tool calls it usually took to get there, and roughly how many tokens that used. 3. Flag the paths where the agent searched, opened the wrong files, or backtracked before finding the right spot. 4. Rank the 5 most wasteful paths by total tokens burned. 5. Suggest the smallest edits to our agent instructions file (CLAUDE.md, AGENTS.md, or whatever this repo uses) that would send an agent straight there. Pointers, not paragraphs. Don't change any files yet. Show me the report first. Make the edits, and run it again. -- Thank me later 😉

After a few days of reflecting on—and dissecting—everything that happened with CLARITY, one thing stands out as remarkable: the tireless work of dozens of staffers who cared deeply about getting this right. And among them is a small group who went above and beyond. They deserve our deepest gratitude. Some of the hardest parts of legislating happen between the headlines: endless drafts, competing priorities, hard red lines, and brinksmanship. That back-and-forth isn’t failure. It’s how serious bipartisan legislation gets stronger. Staffers rarely get the credit they deserve for that work. They put in countless late nights, wrestled with incredibly hard issues, and showed up every day to represent their Senators and their states. Public service is too often thankless. So thank you—to the staffers, and to everyone who read the text, engaged constructively, and called their Senators. Hope we can all keep showing up. That’s how we get clear rules for crypto.

Tesla’s FSD driving is having a “ChatGPT” moment right now. It’s reached a point where it’s actually a better driver than a human. And once you experience it, you can’t have otherwise. Every other auto make is about to see FSD success, and will begin to triple down on advancing their own system. The problem is? They’re ~5 years behind Tesla.

No joke, I spent 20 minutes this morning teaching how to manage your agents and context windows the right way. All of it was manual. All of it took a lot of explaining. Claude's new Projects handles it for you automatically. Beyond that, this drop is huge, including a brand new multiplayer doc experience. @claudeai is coming after every app, and this round of updates made that vision a lot clearer. Tap in and I'll break it all down 👇 0:00 What launched today 0:54 Chat + Cowork are now one Claude 1:29 Claude Design, built in 2:04 Claude Docs: multiplayer docs with Claude 3:07 Comments + Slides 4:28 Apple Notes + Messages connectors 5:11 Scheduled tasks in the cloud 6:19 New settings: artifacts, memory, notifications 7:16 iOS simulator + computer use 7:57 Projects: Claude runs your threads 11:29 Why Projects fixes context management 11:58 Where Claude is headed 13:00 Small business plugin

I'm hosting a virtual workshop with @AnthropicAI today at 11:30am EST (9/17) We'll be covering Claude Code 101: - Preparing your repo - Context engineering and management - Skills / plugins / chains - Loops / goals - Dynamic workflows Generally, a great introduction to agentic engineering + an internal look into how @tenex_labs uses AI tools like Claude Code to do 10x more. RSVP for free here: Hope to see you there!

I can't say too much... But @AnthropicAI has found their groove, again. I think we can all agree that it's been a quieter couple of months from a product development standpoint, but it finally feels like months of their work is now live or about to go live 😉