Code is free. Winning is not.

I knew very little about building mobile apps. Yet in the last 3 months, I’ve built more than I could have with a team of 3–5 engineers pre-AI. A year ago, this was just becoming possible. 3 years ago, it was not even conceivable.

If you’re building software today, you can develop at the speed of thought, and sometimes even faster. You can have a high-level idea of what you want, have the AI think through it (yes – they now come with a “thinking” feature), and then plan <-> implement relentlessly. What previously took a team of 5 or more software engineers could now plausibly be done with 1 person orchestrating 5 AIs who never stop working.

In fact, it’s not a mere multiplier of going from 1x to 10x productivity. In a more important sense, it’s a 0 to 1. A person who could previously never build a product before can now actually build products that others find useful, and do this much faster than if they actually went through the process of learning & building everything themselves unassisted.

So when you’ve got technology that rapidly lowers the effort involved in creation, what’s going to matter?

The 3 things that I think will matter are agency, taste, and attention. These always mattered, but the amount by which they matter in the immediate future will be way higher. I’m not sure if this will continue to be true a decade from now, but I’m reasonably certain that it will be for at least the next couple years – here’s why.

1. Agency

Last month, Base44 got acquired for $80m. This company started from zero, scaled, got hundreds of thousands of users, stayed profitable, and got acquired – all in a 6 month period. And for 5 of those months, it was a team of 1.

The founder, Maor, did not have it easy – he still had to market the product, understand what people wanted, scale, fix the million issues that showed up. But the point here is that that outcome would not have been possible even if you did all those things in the past.

Agency was always more important than intelligence, but intelligence mattered in the past too. But in a world where the cost of intelligence is exponentially dropping, agency will end up becoming much more important than it was before.

You can now pick a problem to solve, invoke an arbitrary number of AI instances, each of which work through different pieces of your product – all planning, building features & fixing bugs simultaneously (while being 10x or more cheaper than hiring an engineer and also being much faster).

To give you an example, back in April – Claude has read through 100m+¹ tokens (roughly equivalent to 75 million words) in my codebase alone. For comparison, it would take 20 years for a person² reading 1 hour a day to consume an equivalent amount of information.

2. Taste

Taste will matter because when you can create anything, being able to perceive what people want, what problems they have becomes much more important.

In a world where everyone has access to the same tools and intelligence, what wins will comes down to who can make something people want.

As history shows us, you can have access to the best engineers in the world and still create products that fail.

Again – taste always mattered, but it does more so in today’s world where the effort required to build a product drops – where we’ve got 273 AI code assistants competing today versus 3 competitors. And that can only come when you really sweat the details and understand what people want.

But it’s easy to get lazy when AI can do the work. If you let AI determine the how andmake all the design decisions, your product will look exactly like thousands of others built with that tool. I’ve caught myself doing this – accepting whatever the AI suggests, then realizing the feature feels generic.

3. Attention

Attention may be the most important thing after all. By attention, I mean the attention of people. We already scroll through more information than what we can retain, and this is going to get much more acute with AI-generated slop (turns out the slop is actually getting pretty good). You can see this with realistic-looking AI generated videos trending on the daily in TikTok.

In that world, capturing a sliver of people’s attention will become increasingly important. Because you may build the best product, but if you can’t get this in front of potential users, the product is dead-on-arrival.

This is why distribution becomes very important – you already see this with influencer marketing on socials, building in public, quirky product launch videos by companies.

If you’ve followed Cluely and how they rage-baited people with “cheat at everything” or friend.com when they acquired a domain for $1.8m (when the total funding they raised was $2m), you already know what this means in practice.

The tools are getting so good that soon everyone will be able to build anything³. When that happens, only three things will separate winners from the noise: the agency to actually ship, the taste to build what people want, and the ability to capture attention in an infinite sea of products.

1. 1

   The sample efficiency and amount of context humans can hold in their brains is orders of magnitude better though – we don’t need to read the same thing over and over every time we work on it. Or the fact that we can retain the information equivalent of more than 2 Harry Potter books in our brains (which is actually the limit of Claude’s models today). But remember that the amount of context these models can hold has grown 200x in the last 3 years.

2. 2

   The average person reads 200–300 words per minute.

3. 3

   This post revolved around building software products, but do realize that this is going to be the case for every domain. Software is where it’s presently apparent because that’s what these AI labs primarily need to accelerate to build smarter AIs. But it refers to “smarter” in the general sense – meaning AIs world class not just at software engineering, but at EVERY cognitive task (marketing, legal, finance, biology, transportation).