It was very surprising to me when I got into venture capital that a lot of my peers don't use — and have never used — the products they invested in.

In one instance I was trying to use a piece of software from a growth-stage company that had stagnated for years. Flat growth. It was one of the worst user experiences I'd ever had.

It didn't surprise me in the least that their user growth was flat.

Now, you can argue that asking investors to be users of every single product they invest in is extreme and unrealistic. Should medtech investors use the neurosurgery products they invest in?

Fair point.

This is why I have huge respect for the consumer AI team at Andreessen Horowitz. Every day on social media you can see them trying out and posting about AI products — ones they've invested in and ones they haven't.

Yohei Nakajima is another person I noticed early on: investor by day, builder by night. I wanted to be him. But I couldn't, because I was not technical.

And then agentic coding arrived. Now I can be.

A startup I've been following, 21st.dev, put out a new agent deployment solution — for putting Claude Code-based agents anywhere you want. With the help of Claude Code itself, I had one deployed to a dedicated frontend with UI/UX components that also come from 21st.dev.

All done in 20 minutes.

Then I spent an hour having Claude Code walk me through the codebase and explain what was happening on the infrastructure side and why it was designed that way.

This changes the tenor of my conversations with startups. Lemme give you an example.

For my portfolio company Gamma, I wanted their API and MCP server to interact better with my workflow, so I could more flexibly build customized decks and one-pagers directly from my CLI. In the process of building that, I learned that JSON config was how Claude Code was talking to Gamma.

About two weeks later, I met a startup that was perceived as building traditional SaaS — a no-code content management system for endpoint displays in retail locations. They'd built this years ago, and at the time they landed on JSON config as the key layer between users and edge content. When I heard this, I blurted out: "That's actually pretty suitable for agents."

The founder looked at me and said, "Exactly."

Now, this is nothing to be proud of. Anybody with a technical background would have recognized that.

But remember: three months ago, I had no idea what JSON was. Or what CLI, bash, or shell was for, for that matter. It's only because I'm building with this stuff that I have a "feel" for it.

Sometimes I have a strong feel for it and I don't even have the language to put it to our investment team (and yes, they rib me for it all the time).

Anish Acharya from the a16z consumer AI team made a point on 20VC: when three new models drop, he's trying all three that same day. That's the bar. Not reading about them, not following the Twitter commentary — actually running them.

An investor I know and respect is in medical device. He invests in heart surgery devices. No, he doesn't try it himself. But he spent three years observing surgeries before investing in a solution (and yes, he made a lot from that). Same principle: if you invest in a category, you have to get inside it.

Anecdotally from around me, a lot of investors aren't doing this. It's hard work. It costs money — premium-tier tools require subscriptions, and most people won't pay. That's actually a filter. If you're willing to pay and use and build, you're already operating at a different level than most of the market.