Is AI Breaking the Lean Startup Playbook?Patrick Collison × Harj Taggar · 主题引导式共读笔记

Harj Taggar: Your most impressive achievement I would argue was Croma, your dialect of Lisp.

Patrick Collison: Compilers do it for us. We don’t mourn it too much. And so maybe in the same way we shouldn’t mourn source code.

We should just transcend the plane of instructions to Claude et al. But emotionally, I miss it.

Patrick Collison: My model of this is cache, the C-H, not an S-H, where Jeff Dean has this famous set of numbers that every programmer should know—bandwidths and latencies and just kind of relevant constants that you should reason about as you build systems.

Retrieving something from L1 cache is very different from retrieving from RAM, which is very different from retrieving across the network or whatever.

That’s a hell of a lot slower than knowing it in cognitive L1 cache.

And so I think even granting the full capabilities of the models, I still think there’s a pretty—I think for a long time to come, neuronal lookups will be much faster.

If you look in revealed preference at what companies themselves are doing, whether they’re companies like Stripe or the labs or what have you, there still seems to be an enormous premium on cognitive ability.

I think renouncing that before there’s evidence that we’ve saturated those benefits would be premature.

Patrick Collison: I still write myself. I both philosophically, but also specifically, substantively dislike the writing of the models.

It’s very interesting, right? Because these can prove the Jacobian conjecture, whatever. And so clearly they’re capable of these monumental feats.

But somehow I still haven’t read the LLM essay that I’ve found super compelling.

It’s just very hard to RL in that domain because the utility function or something is kind of hard to define.

But yeah, I think writing is pretty—interpersonal communication and writing I think are so very fundamental and being able to reason sensibly in the multidimensional space of reality.

I felt this real sense of urgency, which I think in hindsight was a bit unnecessary.

I thought that a bunch of the opportunities in startups and in Silicon Valley and so forth were ephemeral and fleeting.

In hindsight, I think that was a poor intuition. It’s been pretty robustly and reliably the case over many decades that Silicon Valley has a surfeit of opportunities.

Humanity has always had an affinity for these millenarian models of how everything will soon come to an end and be this permanent transformation of society and so forth.

Obviously, aviation was a pretty big deal, but I don’t think it was quite the sociological rewriting that some of the excitable proponents at the time imagined.

So it’s hard to predict anything, especially of the future, but I would take the under on this being the last couple of years to create a company.

It’s very easy to hallucinate or to imagine some customer problem that’s not actually something viscerally felt by a person who would pay money.

We felt like the proverbial squirrels in a trench coat trying to masquerade as a real business or as serious adults, but obviously knowing nothing coming in about the space.

But I think the fact that it was grounded in such a concrete, actual, real user problem saved us.

I think the thing that saved us and meant that it wasn’t a total walk in the wilderness is we had production users almost from the very beginning.

We got our first live production user in January of 2010, so two months into working or whatever. And it did very little. It was very larval and incomplete.

It was very just-in-time development.

Every week we had actual customer feedback, requests, new users coming in; we’re learning things from reality as opposed to our own hypothesized or extrapolated conception of it.

I think if you have a significant stream like that of grounding, it’s probably okay to not be launch, launch.

Taking these really divergent starting points where nobody else is trying to occupy that territory is maybe more—basically, maybe you have to more aggressively de-correlate in the era of AI.

So many of them are very anti-lean startup, right? Whether it’s the labs themselves or Anduril or you can go down the list, a lot of them have this characteristic.

Whereas now I think you can start these much more aggressive and ambitious things upfront.

Harj Taggar: Within YC and probably the startup world at this point, you’re famous for at least the Paul Graham term, schlep blindness.

Patrick Collison: I think PG latched onto something where there are all these menial tasks, but in the totality of Stripe, I find it so interesting.

Every business is an applied theory on how some aspect of the world works or how some market works, or if it’s a new company with a new model, that’s a contrarian thesis on some counterfactual.

Will rapidly improving AI capabilities do this or will the labs specifically themselves do this?

Human organizations are complicated and it’s very hard to aggressively prosecute 100 different priorities and to deal with all the issues and interference that arises among them.

Google has done incredibly well in a bunch of specific places, but it’s not like Google has done all the things, even if in some basic material sense, Google maybe had that ability.

So I’d say that the track record of that is checkered.

Just literally LLMs will obviate a bunch of, or agentic capabilities will obviate a bunch of specific verticals or tasks or something.

Harj Taggar: Within the YC part of the life cycle, like day zero to 90, it’s really being driven by enterprises willing to buy from startups, which is the new thing. So you can sign these new contracts within the batch.

Businesses everywhere are more spring-loaded to adapt and to try new things.

They have a real terror of being left behind with archaic and antiquated ways of operating.

There’s never been a better time for startups to sell and to have their products get adopted at pretty meaningful scale right out of the gate.

There’s a fear that AI is going to be this hegemonic, centralizing, totalizing force where a small number of companies gobble up a very large share of the economy.

I think there are going to be many thousands of winners. Based on the trend lines we can see, I think we are heading towards a more decentralized world and one with more broad-based prosperity.