Dave Caplan

LLMs are a virus, they have infiltrated every corporate level on a global scale. This is the so-called “singularity” and the final boss of globalization. Everyone using the same borg brain to “think” and make decisions. Thoughts? Anyone?

Most frustrating thing I’ve seen today: @intercom website has no login button nor does it even have its own support widget or link to a support page. Maybe I missed it though

Honestly, traditional visually appealing web design might soon be a thing of the past. We need to start building websites that LLMs like and can quickly navigate. Is it just markdown? Or will web accessibility standards (for screen readers, etc) soon enter the spotlight?

Ok we get it, AI is cool. But have you tried

If you ever want to be humbled by AI type /insights into Claude

I’ve looked inside a few engineering orgs recently. Mostly teams inside traditionally non-technical companies. What shocked me wasn’t bad code. The code was just fine. Tests fine. CI/CD also just fine. It was how little structured AI leverage there was. Everyone dabbling. No system. No multiplier. There reason is obvious and not at all what you might think.

Gotta get that one last prompt in before bed

Even if AI takes over the world, at least we’ll still have Valentine’s Day… right?

2009: “First.” 2016: “Fake news.” 2026: “Is this real?” comments are evolving

Anyone else’s Claude just get tired at the end of a long day?

Can you imagine 1609 when Shakespeare dropped Sonnet 5 on the masses absolutely ruining every other poet overnight Anthropic 🤝 history repeating itself

Claude Code just added teleport. You can now build and ship from the toilet. Society is not ready.

Who else is vibe coding AI tools for investing? So far I’ve built agents to manage options wheel cycles, rank upcoming biotech catalysts, propose debit spreads and lotto calls for biotechs, and a bunch of other things. Best so far has been the wheel agent but you don’t really need AI for that. Everything else is still as unpredictable as you’d expect.

Everyone’s talking about getting rich vibe-coding small apps, but here are some bigger bets to take this year if you want to create something truly valuable: 1. “ChatGPT / Claude for X”, where X is off-limits to general models (privacy, security, regulation, proprietary data). Needs to be as good or better than the base model. Example: OpenEvidence 2. Things that make Claude meaningfully better for a specific vertical. Claude Skills are the real unlock: packaged workflows + domain expertise = massive output gains. Expect marketplaces to form around this. 3. Products that help humans talk to other humans. Dating. Conversation. Coordination. Connection can't get automated away and only gets more valuable as AI eats attention. Example: Twos 4. Conversation-based tools for specific tasks. Phone or text. No UI. No app. Just results. Example: Boardy (love it) 5. Infra that makes AI usable in the real world. Reliability. Security. Observability. Shipping across environments. Anything you build purely for yourself to automate, learn, or make money? Don’t bother scaling it. Anyone else who cares already built their own. And none of this changes the fundamentals: trust and distribution are still king. I wrote this as a reminder to myself. Hope it helps you too.

2026 is when the app era actually starts dying. ChatGPT and Claude didn’t win because they’re smart. They won because people don’t want to do shit anymore. They want to say what they want and have something else do the work while they live their lives. Clicks, menus, dashboards, onboarding flows... all friction nobody needs. Conversation will beat UI (almost) every time.

I love side projects. Worst case: you learn something. Best case: you accidentally build a company. Happens more than people think.

What are the best tools people are using these days to teach Python or JS to beginners? VSCode? Jupyter? Replit? GitHub Codespaces? Google Colab? Glitch? I’m thinking short feedback loops, good UX, etc..