How I build 10x longer on the same Claude plan

Anthropic's Opus 4.5 dropped a few days ago and it's incredible, but people are complaining they're hitting limits within like an hour of using Claude Code. Here are three settings that let me code all day long without getting throttled. I'm Pete. I own a software development agency in New York City. We work with startups and Fortune 500 companies to bring their products to market. Over the last few months, I've been building my dream social network almost entirely in Claude Code, and I never hit limits on the Max plan. Here are my three tips. First, stop using the Opus model exclusively. Not a lot of people know about this. Use the Opus plan instead. Now when you're in planning mode, Claude Code uses Opus for strategy, and when you're executing on the plan, it automatically switches to Sonnet 4.5. Sonnet 4.5 is a really capable model. I've been using it for almost two months, so you get Opus level reasoning where it matters most without burning tokens on the implementation. Second, keep your CLAUDE.md file lean. That file loads and sessions start and it stays in your context window the whole time. The bigger it is, the less room you have for actual work. And make sure to keep it under 40,000 tokens. If you go over, Claude code will actually complain that you might experience performance issues. And also watch my video on cascading context to make sure Claude Code has what it needs while staying trim. I also have a substack article on this topic. There's a link in my bio. And finally, install AST grep. So Claude spends most of its tokens searching your code base. Ripgrep searches text, AST grep searches structure. For example, if you're looking for all async functions without error handling, AST grep will find them accurately in one pass, the way I use it. I installed the Claude Skill from Github and Claude will use it automatically. It's a massive token savings.