The best way to figure out what works is to ship something and find out.

What a time to be alive. On a plane, posting to social media, watching the Super Bowl, working on a MacBook Pro, collaborating with AI, and managing projects semi-automatically with Linear + MCPs. This used to take a year; now it's just a plane ride.

When benchmark manipulation is the strategy, trust is the casualty.

BetaKit's 30 AI predictions for 2026 are missing the big one: AI's real bottleneck isn't chips or electricity. It's trust. The hardest problems require the most vulnerability. And you're not sharing your deepest context with a system you don't trust. betakit.com/thirty-bold-pr…

Menopause, long covid, divorce, career upheaval. These aren’t Google searches. They’re journeys.

The wealthy have always had access to top experts. Now the rest of us want in.

AI that removes humans from the loop removes the thing that made it valuable.

Since Spotify went AI-driven, I’ve been listening to more full albums and building my own mixtape-style playlists. It’s been great. There’s a real sense of nostalgia in it: sitting with an album from start to finish, following the journey the artist meant to take you on.

Vercel’s new Skills ecosystem for agents is cool. Also a supply chain security nightmare waiting to happen. Natural language instructions get ingested directly by LLMs. No static analysis. No sandboxing. Just vibes and hope.

Every intimate conversation with ChatGPT is now optimizing for ad engagement. They didn’t sell your data. They built something to exploit it.

The hardest problems require the most context. That’s why generic AI keeps hitting a wall.

Everyone’s building AI to replace experts. Almost nobody’s building AI to amplify them.

the companies racing to remove humans from the loop are building products nobody will trust with anything important

OpenAI is asking contractors to upload actual work deliverables from their jobs. Not to learn from them. To benchmark when AI can replace them. These contractors ARE experts. And they're being paid to make their own expertise obsolete. 🤦 wired.com/story/openai-c…

If you have a mind, you’re a genius. That’s not motivational. It’s definitional.

generic AI advice is worse than no advice because it gives you false confidence.

Without trust, you get surface-level questions. With trust, you get the real problem.

Scaling an expert is valuable. Replacing them entirely defeats the purpose.

new claude code feature: add ! to a prompt for it to be run as an in-context bash command 🔥 Finally! I don’t have to keep two terminal windows open anymore for basic stuff.