Dark Startup Factories

Let's talk about startups.

When a company finds product-market fit - usually what you need to do is to double down on it - you push the gas pedal to the absolute bottom and you do everything to grow your company and the product as fast as possible.

Traditionally the only limiting factor was just how fast you could ship - which was limited to how fast you could hire, which was limited by how good you are at attracting talent and managing people + company culture.

Because your resources are limited, it is also crucial to have a good product navigational skills. Knowing where to steer your precious development time. Not to dive into unnecessary features and instead develop features that give you the biggest leverage.

Agentic startups

In the era of agentic coding, this might not be the case for much longer.

(For the sake of argument, I'm assuming that coding models continue to improve to the point that they are roughly on par with humans and the cost ends up being at least roughly on par with human cost.)

As soon as you hit on some sort of product-market fit, the only limiting factor is how much cash you have, and how fast and efficiently, you can burn your tokens.

So before you had a system where you had a human-bandwidth constraint, but with agentic startups, you are compute-throughput constrained.

This removes a lot of friction from startups but introduces a lot of fundamental changes

• Having an architecture that allows you to scale your code gets way more important than management
• Your product sense actually becomes more important than before, since it's much easier to build your backlog
• Your people skills are almost negligible - outside of being able to sell a vision and talk to VCs (arguably, that was the case already)
• Historically the best startups have been started in the Bay Area, since it has the most abundant pool of talent. Once that's no longer the limiting factor, the ideal startup's location is constrained more by data privacy laws and access to capital.
• Runway used to mostly be your cash / monthly payroll, but now it's just a matter of how large the features are that you're building at any given point.

Let's talk about runway

The Runway paradox

Your runway used to be a super important metric for a startup. In order to build more features, you have to hire more people. Scaling down is really hard - it's tough to fire people, it sends a bad signal out (you either hired too much or you are failing - both are bad) and nobody wants to work at a startup that's already firing people.

Traditionally, stretching your runway gave you time to find product market fit (if you were in the pre-PMF phase, that is).

If you don't have to hire that many people in the first place, your runway becomes more ephemeral. You might spend 90% of your runway in 6 months, and then just sit and watch the effects.

When you think about it from a VC's point of view, you might even start demanding that your startups just spend all their money immediately and make the product as good as possible in a month, and see if it takes off from there.

You could even just pour $100k in immediately, measure, and then immediately kill the idea - or if it's working, then dump additional $1M .. and so on.

In any case, the idea of a Runway becomes less relevant.

Do we need startups?

As models become better, we'll also start seeing additional changes. The humans in control might not be making the best decisions to begin with. They might be good idea incubators, but perhaps not the best operators.

Humans are also slow, emotional and prone to all kinds of fallacies. If scaling is a math problem an agent can math much faster than a founder looking at an analytics dashboard.

From a growth-perspective - there's perhaps a pretty linear path to take from product market fit to scaling up features and the product - you might not need the humans there in the first place.

It used to be that you need good operators, to scale up the product and grow a team. Having a product vision driving you, driving your employees - those were important to keep people motivated.

But if there's no people to motivate, do you need founders?

Then the question becomes "why should VCs invest in founders?"

In theory you could just build an idea factory, perhaps with a human in the loop, and they just start pumping out ideas, and putting each idea on a factory conveyer belt. Where each one gets processed and refined and tested by a thousand different agents. Like a Dark Factory.

Let's call that a Dark Startup Factory.

Are we there yet?

We aren't. At least the agents are not fully able to operate autonomously without producing something that's hard to maintain - even for them. Without a human in the loop, code tends to degrade to unmaintainable spaghetti. Coding models are also pretty bad at design and understanding what makes people tick, so there certainly are limitations.

At the moment.

But eventually the LLMs will be better than us - if not from a qualitative perspective, at least from an efficiency perspective. Dark Startup Factories will be built.