At USV, we have been spending a lot of time asking the question of what we would look like  if we built the firm from scratch today in an AI-native world. What’s our new OS? The answer is that many of the core principals stay the same: small funds, tight partnership, and a thesis-driven model with specific points of view (the more specific the better.) We still believe in a model that prioritizes relationships with founders, teams, and with each other. That will not change.

But we also think a lot can change and quickly trying new things to figure out what that might look like has been a lot of fun and fruitful. The speed of thinking of something we can try internally to getting it into our workflows feels awesome every time. That’s why we’re doing things like rethinking our entire internal operating stack, pushing you animated cartoons of Fred as a piece of bread, doing weekly internal show and tells of things we’ve each played with or built for ourselves, hosting “New OS” demo days, and hiring a designated AI lead onto the team. AI gives us new ways to organize, synthesize, and connect information. It helps us sharpen ideas, cut through noise, spot the right companies sooner, and manage the machinery of a fund more smoothly. Board reports, LP updates, reserves management, meeting notes—all the often unglamorous but essential work of a partnership—suddenly has leverage. Tasks are faster, leaving us more time to spend on what we believe is going to change the results most significantly. And, at its best, its not only about saving time but getting closer to the weird, the edge, the unpredictable. That’s where we like it best. It’s been exciting to see the range of what new tooling can do for us but also interesting to understand where the constantly shifting limits currently sit. 

One thing we have learned is that the real gap between humans and AI is not usually in the facts but in the takeaways. A small example is board-meeting reporting. Collectively, we sit in about 300 board meetings a year (that’s a wild guess but seems directionally right.) For years, our process was simple: circulate the deck with a partner-written summary on top. The summary always had a few parts, the facts (we focused on X, the biggest challenge is Y, growth looks like Z, check out slide 12) and the takeaway (I am optimistic because the team feels tighter and more prioritized than last quarter, I am concerned because there is visible tension on the exec team, and so on). When we moved these write-ups to AI, the facts came through fine. It was definitely faster, easier, got the job done. But the takeaway, the judgment, the vibe, was missing. Reading them felt flat. Over years, we have each learned how to read the tone in one another’s notes and to catch the signals in someone’s phrasing. I missed the quick hot takes that I'd come to look for that always said far more than the summary of the data. Now we use a hybrid. AI handles the deck, the notes, the bullet points. Each of us adds a human TLDR at the top, something like a vibe report. It takes a minute or two more but the mix has been far more useful (both for the reader and the person sending around the report--that moment to capture the real point has value in it, too.)

The same pattern shows up in our thesis work. On a recent LP call, one of our investors asked: with AI, what is table stakes and what is alpha? I liked this framing. Efficiency gains like custom GPTs for portfolio tracking or automated reporting are quickly becoming baseline. Like email or Slack, they will soon be assumed. The fact that you use them is not interesting. Where the alpha still lives, at least for now, is in the quality of the ideas you put into these systems. AI can take our theses and get them out faster, push our outbound further, and process diligence deeper. But without sharp direction, it just leaves you in the middle of the pack. With it, we can use the tools we now have access to and figure out how to chart our own course. The leverage is real, but, at least for now, realer if the human input is the differentiator and the tools are the accelerator.