Seed VC: When to Use AI, and When to Use Your Spidey Sense

My best thinking happens in the shower or on the walk to the subway. Not in front of a screen. And definitely not before or after opening an AI tool.

At the screen I’m doing research, consuming decks, looking at spreadsheets. It feels productive. But it’s not really creative thinking. So much of VC is the thinking process, not the research.

A lot of times my colleagues ask what I saw in a given company or founder, and I can’t fully articulate it. That’s kind of the point. Pattern matching, “spidey sense,” or just having spent real time with an entrepreneur is what I’ll say. It feels incomplete.

Fast forward to now. I use AI constantly. I’m not making a case against it – it’s changed my workflow forever. But I’ve started to notice what it’s actually good at versus what I still need to do myself.

It’s a remarkable analyst. Market history, regulatory landscape, comparable exits, how a category evolved. It synthesizes fast and well, and over time it’s really learned how we apply our framework, though it often takes our principles a bit too literally. Work that used to take days now takes minutes.

What it can’t do is tell me what I actually need to know at seed.

So much of the job is figuring out the “why.” Why this person. What’s the narrative behind them. What have they lived or lost or built that makes this the problem they were always going to solve. That stuff isn’t online. It hasn’t been written yet. It exists in a conversation, or in the way someone answers a question they weren’t expecting, or in a reference call where someone pauses before they answer.

I wonder how AI would have scored Uber in 2010. Regulated industry, entrenched incumbents, impossible go-to-market, TAM framed as “replacing taxis.” The real market – replacing car ownership, transforming logistics – wasn’t in any dataset. It didn’t exist yet. The model would have flagged every real risk and missed the only thing that mattered.

I passed on Doorbot, which became Ring and sold to Amazon for $1B. I didn’t live in a house. Couldn’t feel the problem. AI would have agreed with me. “Niche use case, limited TAM, why wouldn’t ADT do this.”

Raj De Datta recently posted that AI has actually increased his cognitive load, not decreased it. He’s running more projects in parallel, context switching constantly, doing harder thinking – he just no longer has to do the easy thinking and doing.

I feel the same way about investing. I have more information. Better research. Faster synthesis. And I’m genuinely not sure I’m making better decisions because of it. Maybe better-documented ones.

The shower is still where the real work happens.