Anthropic just dropped Claude Opus 4.8. And what comes next is worth paying attention to. Not because it's a shiny new model. Because the gap between what's available to you and what's being tested behind closed doors is about to close fast. Here's what actually matters from yesterday's release: Opus 4.8 brings real improvements in coding, reasoning, and agentic task performance. But the headline isn't the model. It's the signal buried in the announcement. Claude Mythos Preview, the most capable model in Anthropic's stack, has been running in limited access under Project Glasswing, a controlled cybersecurity testing program with select enterprise partners. The reason it hasn't been released broadly is straightforward: the model's capabilities require stronger safeguards than have been fully developed and validated. That's not spin. That's Anthropic being honest about the responsibility that comes with capability at that level. Major tech firms including Amazon, Microsoft, and Apple are currently permitted to use Mythos for cybersecurity purposes under Project Glasswing. What does this mean for you as an entrepreneur using AI right now? The tools you're building your systems on today are not the ceiling. They're the floor. Opus 4.8 can already run hundreds of parallel subagents in a single session, carry out large-scale code migrations, and plan and verify complex work with minimal human oversight through a new Dynamic Workflows feature currently in research preview. Most people are still copy-pasting prompts one at a time. The entrepreneurs who will win the next 18 months aren't the ones with the best ideas. They're the ones who built systems before the capability wave hit. Anthropic also introduced a fast mode that runs at roughly 2.5 times the speed at three times lower cost than previous versions. The cost excuse for not building is disappearing. The window to build before everyone else figures this out is narrow. What's one system in your business that you're still running manually that AI could be handling right now?

Most people pick an AI income model based on what sounds exciting. That's why they stall. The right model isn't the most popular one. It's the one that matches where you actually are right now. So before you pick anything, run yourself through this. Start with how fast you need money. If you need income in the next 30 days, your two best options are SMMA and AAA. SMMA stands for Social Media Marketing Agency. You run content for businesses that don't have time to do it themselves. Your offer is simple: one monthly call with the client, you record it, feed it into Claude, and out comes 30 days of posts and carousels. You charge a monthly retainer. They get consistent content. You get recurring revenue from a process you build once and repeat. AAA stands for AI Automation Agency. You build workflows and automations inside client businesses. Connecting tools, eliminating manual tasks, building AI assistants that handle intake or reporting. Clients pay well because the pain is real. The catch is every client's system is different, which makes it hard to scale without a team. SMMA gets you to a paying client faster with a lower technical barrier. AAA has a higher per-client ceiling but requires more skill and more patience. Both are solid starting points. Neither is a long-term empire on its own. If you can wait 3 to 6 months, more models open up. AEO agencies help businesses show up in Google and AI search results. Vibe coding agencies help non-technical founders finish and ship the product they've been stuck on. Information products let you package what you know and sell it at near-zero marginal cost per customer. If you're playing a longer game and already have some income stability, communities and SaaS are worth building toward. A community at 200 members paying $50 a month is $10k MRR. SaaS is the highest leverage of all but the slowest path to first revenue. Not for someone who needs money now. Next question to ask yourself: do you want to sell your time or build an asset? Coaching, consulting, speaking, and agency work all sell your time. Some of it's repeatable. None of it fully disconnects from you. Courses, communities, and SaaS build assets. They take longer but the ceiling is different. Neither is wrong. But you have to know which game you're playing before you start.

For most of my life I thought the problem was motivation. If I just wanted it bad enough I would do it. So I kept trying to want things harder. Didn't work. Turns out the issue wasn't desire. It was executive function. The part of the brain that initiates, organizes, and follows through. And when I started using AI to fill that gap, not to replace my thinking but to support the part of my brain that struggles most, everything got easier. Not perfect. Easier.

Boris Cherny: "Claude Code is 100% written by Claude Code, Cowork is 100% written by Claude Code."

Here's the framework I use to evaluate any new project before I commit: I call it the Four-Gate Filter. Gate 1: Is this interesting enough to sustain attention through the hard parts? ADHD brains need interest, not just importance. If a project only matters without being interesting, it won't survive contact with difficulty. Gate 2: Does this align with where the business needs to go in the next 90 days? Interesting but misaligned is just a distraction with better aesthetics. Gate 3: Do I have or can I build the system to execute this without depending on inspiration? If the project only works when I'm motivated, it won't work when I need it most. Gate 4: What does the first three steps look like? If I can't define the first three steps immediately, the project isn't ready to start. Projects that pass all four gates get a start date. Projects that fail any gate go on the someday list or get dropped entirely. This single filter eliminated most of my unfinished projects. Not because I stopped starting things. Because I started starting better things. What filter do you use before starting a new project?