More Waves Than Ever

It’s been a while since I posted.

Partially because I’ve been busy. Partially because I was trying to find the right angle for what I wanted to say. Then last week I spent a few days at TechCrunch Disrupt in San Francisco, and it felt like the right moment to finally talk about innovation.

Faster, Easier, More Expensive

AI is enabling us to build faster and easier than ever. As long as we are willing to pay the ever-rising price of tokens.

Yet building the right product is becoming harder. And even if you build the perfect product, finding product-market fit is insanely difficult and volatile.

After all, you are but a drop of water in an ocean that is generating more waves than ever.

The Founders

I talked with many founders, many of them pitching super interesting products. Coding harnesses leveraging all sorts of techniques and concepts I keep reading about, from Karpathy’s LLM wiki to recursive self-improvement networks.

All of it with a blend of pricing. Subscriptions, the ARR model. Or the old European model: buy the license, then pay the never-ending consulting fee to adjust the product into something that vaguely resembles what you need. Been there, done that, learned a few lessons.

It was great to feel the passion of the founders. It re-ignited my spark of building something.

Signal Among the Noise

Not everyone who reached out wanted to sell me something.

Some people just wanted to connect. I remember a great conversation with someone thinking about how to turn decluttering and selling their stuff into a business, something I’m familiar with. And a genuinely candid young engineer finding her footing in this hard and always rushing industry.

It is great to have the opportunity to listen, learn, engage and, in a few cases, offer some words of feedback or my own thoughts. It is nice when the context allows you to do this freely.

Everyone Is Reading the Horizon

I listened to all kinds of talks. How to tell if you really have PMF, with Rahul Vohra, who wrote the classic on it years ago. How to get your first 1,000 users without a marketing budget. The big VCs explaining how they are thinking now about investing in new ideas.

Crystal Huang from GV was brilliant. Her take on stickiness is worth reading: if your product is easy to implement, it’s just as easy to uninstall. When anyone can build anything over a weekend, that one stings.

My reading: everyone is struggling to figure out where we stand, and trying to read what is on the horizon.

Many talks were about moving to open-weight models. That hints at people who have been around the industry for a while. They know that in a gold rush, the ones selling the pickaxes are the ones making money.

At one roundtable on open weights, someone said something I remember vividly, more or less: I don’t need a frontier model able to translate from English to Aramaic for my use case. It makes sense. As long as you are also able to secure the infra to run your workloads, either in the cloud or on premise.

Then Jev entered the landscape, and within days there was a bunch of clones. Interesting idea. Still figuring out the applications.

And now there are echoes of LLMs being doomed. An old paper on how scaling works is making the rounds again, YouTube is buzzing about the limits of scaling, and the frontier labs keep stretching those limits, release after release. Meanwhile, a new player in town. Looking at you, NCM.

The Unpopular Talk

And then there was the talk about security. The unpopular one. The one we probably should listen to more carefully.

I’m still digesting it and recalibrating my thoughts, in a week plagued by breaches and attacks run by autonomous agents that are making the industry shake.

The same speed that lets a founder ship over a weekend lets an attacker do the same.

Innovation, in Practice

But I came here to talk about innovation, which is what I have been doing these months.

Interesting buzzword, hard to put in practice. Especially when building is easier and faster, but building the right thing is still complicated.

I talked with two companies using AI for user research. One uses synthetic users. The other uses AI to create surveys that run on actual humans. Very different price models… and I guess very different kind of results.

The Nail That Sticks Out

Innovating is hard. Maybe more so in Japan, where standing out has traditionally been frowned upon. 出る杭は打たれる: the nail that sticks out gets hammered down.

You need to develop a thick skin, and smile while your ideas or intent are misunderstood, criticized, or even openly bashed.

It’s fine. That’s what the founder spirit is about. Real innovation means carving new paths, going against the current, and none of that is easy. It is part of the journey.

A Lot of Rain Since Then

Reading back my last post, I guess there’s been a lot of rain since then.

Now we have Astra, Sol, and many new open releases. I’m particularly fond of Ornith, which released its 1.5 122B model today. Yet Qwen remains king for many cases.

And I’m still rooting for Europe in the AI race. Mistral is a very decent model; I wish it came in a format I can run locally. Yet it’s great to see Europe positioning.

Argon was also released. Good for demos, not so much for serious work as far as I have tested. But it’s also hard to allocate the time and attention to test it properly.

On the bright side, there is good progress in the projects I have been working on and sponsoring. Hopefully I will be able to talk about them at some point.

On the personal side, I started working on my own take on how I would like to work: Polaris. Or should I say, letting Polaris build itself. Not more than a couple of days of prompting. Whether it’s the right product or not, who knows. And if it is, I guess it will be reverse engineered soon with REA, which is also shaking the world.

A drop of water. More waves.

Uncanny Valley, Fresh Air

Also on the bright side, it was fun to ride a Waymo. It feels like the uncanny valley applied to driving. Works OK. If drop-offs and pick-ups can be refined, even better.

I love the weather of SF, not so much the streets. I guess when it comes to cities, it takes time to get used to them. Happens the same every time I go to Madrid, and that’s where I’m from… technically.

Autumn comes. Tokyo becomes fresher, drier, quieter after the summer months. Looking forward to the autumn colors this year.

紅葉 is beautiful.