So... a happy "six after the war" to everyone.

Israelis developed technology that lets aircraft circle Iran like an Independence Day flyover. And still, AI cannot draw a clock showing six o'clock.

While we ran between shelters, safe rooms and Telegram pushes, the world kept running forward. So here we are. Back to building. Thinking. Polishing products.

Meanwhile, Andrej Karpathy gave a terrific talk at YC. Here are the ideas that caught me most when thinking about how to really build products for the AI era:

1. Keep the familiar interface. Leave the sophistication under the hood. Users do not want to relearn.

2. Put the AI behind the scenes. Let it handle the complexity. You focus on simplicity and zero friction.

3. Give control. Give certainty. Let the user choose how much help they want. Do not force them to trust something they do not understand.

4. Build for a new kind of user: AI agents. Yes, your next customer might not be human. Stripe already offers dedicated documentation for agents. Tools are emerging that translate a GitHub repo into text an LLM can process easily. And there are startups being built agent-first from the ground up.

Want to win? Start here: identify where users burn time on manual actions and let AI solve it. Build AI-embedded products without breaking the existing interface. Use AI distribution platforms. Design for the new programmers, the ones who code in English, not just Python. And build not only for humans, but for AI agents.

For anyone weighing how to approach AI products intelligently, we expanded on this in Grove Ventures' Enterprise-AI Playbook.