Attention is all you need, dear startup

A non-AI book scanning machine at the ARC booth at Ai4. How’s that for a business model?

Last essay I wrote about companies looking for control of AI at the Ai4 conference, so have a read if you haven’t already. This time I want to focus on my other observation at the event: why are there so many of these state-of-the-art AI companies already? And why is everything free and also very expensive?

My current refrain these days is that software is a commodity. Anybody can build anything. And, yes, I am aware I’m making a broad statement; I’ve worked in enterprise and understand all of the complexities therein. I often say that system integration is where software projects go to die.

Sotto voce

Case in point, the voice companies that were exhibiting at Ai4. Everyone from ElevenLabs and Deepgram, which are big brands in the space, to a whole bunch of much smaller, more specialized players.

You may have read that ElevenLabs, an early leader in the AI voice space, had an $11B valuation in February (reportedly at $22B now) and more recently, Wispr Flow has been valued at $2B. Yet here I am using Fluid Voice for free on my Mac instead of Wispr Flow and it’s excellent. And, ElevenLabs’ voice cloning technology, that only a couple of years ago was magical, has been replicated for free by Resemble AI’s Chatterbox, or more bare metal, Nvidia’s Magpie and Fish’s Audio S2, none of which I’ve tried yet.

So what’s going on?

ElevenLabs is worth $22B

Let’s look at ElevenLabs, a company I’ve been into for a long time (watch the intro for some silliness). In the early days, they did two things really well: had unmatched voice models that sounded like real humans, and a wonderful UX / UI. All of this is still true. You can have real, human sounding voices, shaped how ever you want, you just pay for some tokens and off you go. They introduced high quality voice cloning relatively early, if not first, and you can still do that today as well. And it is clear and easy to use, and each release blew my mind. But now, as I noted at Ai4, there are a bunch of companies, big and small, focusing on doing all of this, emerging within a very, very short period of time. Some compete on price, some compete on vertical or horizontal expertise (a call center is the most obvious use case), but they are getting harder to distinguish on pure technology.

So how do they win? I think, in the shorter term, by running a real business. They have distribution, notable customers, and enterprise controls. As we all know, building a piece of technology is very different than building a product at scale, i.e., a functioning company.

But to my mind, that doesn’t justify the $22B valuation. Obviously ElevenLabs is going to grow, build more product, expand outwards with new product lines, and so on. And they’re not a SaaS either, it’s pay as you go like an AI model. But I suspect what’s really going on here is that as most of the core technology commodifies and distribution becomes the hard thing, having a huge chunk of customers on your platform and trusting you is all that matters. If you or your organization trusts ElevenLabs and they deliver, who cares that there’s a free model somewhere for cheaper? If they pivot to becoming a full stack customer success tool because that’s where the market is going, fine. If the pricing model changes because we’ve settled collectively on some post-SaaS model (consumption? results-based pricing?), fine, we trust you and get it. Their business model right now may not be their long term business model.

What this means, to me, is that VC has effectively split in two: one funds deep tech, hardware, and clear, defensible moats. And the other has inherited the old B2C playbook where you buy attention, earn trust, and use money as a blunt instrument to own the space. You then trust that a clear business model will emerge before the funding runs out. I think voice startups are a good example of this.

Back in my day, 👴🏻, software was hard; the skill was scarce, unevenly spread and slow to gestate. Distribution has always been hard because it requires money. And, while the current thinking is “distribution is all that matters” (and I agree), it’s also all we’re left with.

Even though ElevenLabs is bringing in revenue (and not even a small amount), getting mindshare and marketshare is what matters, weaponizing VC money to acquire customers like many B2C companies did 15 years ago (Uber, AirBnB, etc.). Also, did you know that Jamie Foxx and Eva Longoria invested in the Series D? This was news to me, but again, this makes the platform seem more trustworthy for creative talent. No one asked me to invest, but I have a shiny $5 CAD bill and my voice is ready to go. Call me.

Free as a strategy

Let’s look at the open source side of things too. I couldn’t understand earlier this year what was going on. State of the art tech was being rolled out by startups that were just as good as the priced solutions, as I noted above, and brand new technologies and products (OpenClaw, Hermes Agent), were just... free. An investor friend of mine suggested that it was the Valley just doing what the Valley does, building the hippie open source tools for everyone to commonly use. I don’t think that’s totally wrong, but as I’m inherently suspicious, I did not believe him and assumed there was a profit motive somewhere that I just didn’t get.

I then realized that open sourcing or sharing the code is, effectively, a customer acquisition channel. Get customers installing your product, have contributors add to it and become invested in the ecosystem, then find a business model later. It could be Red Hat style which is to take an open source product (Linux), and build a service and support business around it. It could be where Hermes Agent seems to be going which is some sort of OpenRouter-style portal but “you should use our platform and because it’s plug-and-play with our tool and we can collect a consumption tax along the way”.

Spoon-fed

You may have heard of Bending Spoons lately as they’ve been on a tear. They recently bought Airtable for $1.285B when not long ago it was worth $11B, and, more recently, Miro for $1.355B against $17B five years ago. They own Evernote, Vimeo, WeTransfer, Meetup, and AOL (!). It’s easy to think they’re just leaning into the SaaSpocalypse and getting while the getting is good. But I see it as getting a big fat user base that’s already been monetized, and then figuring how to better monetize them in the future, possibly even grouping them together in some way Microsoft Office-style. Why start at the beginning when you can start at the heavily discounted late-middle?

It’s also worth being precise about what happened to those companies. Airtable had a business model, as did Miro. Seats, subscriptions, procurement, product-market fit, the whole shebang. The problem for them was that they didn’t last long enough to hit their valuations before they became cloneable.

Chris Mudiappahpillai and I were chatting on the pod a few weeks ago about whether a company is a product or a utility. With a utility, there’s no expectation of innovation, just a bill. So Bending Spoons is perhaps buying products that end up being utilities and prices them accordingly.

While cloning software is “easy” now, what was hard, and still is, is building a company around it and getting millions of users (or big companies) to actually show up. Bending Spoons didn’t clone, they bought the end result of all of that work at a 90% discount from people who spent a decade doing the company building thing. So far, Bending Spoons has been acting more like a private equity firm: buy it, cut costs, raise prices. But I could see them doing a cross-sell / upsell “one login one bill” kind of thing to leverage the customer base they have. No matter what they end up doing, they have the user base.

Buy new, buy used

What’s being bought here is attention, habit, and lock in. ElevenLabs is buying it new, at full price, and blitzscaling as quickly as possible. Bending Spoons is buying it used at a 90% discount, after someone else has paid for customer acquisition.

ElevenLabs and most of the show floor at Ai4 are effectively running B2C playbooks inside B2B companies, which is very different from where software startups have been over the last few decades. You used to build something hard and that bought you a defense. Attention isn’t a moat, which is why Coca-Cola has never stopped advertising. They stop that, they stop existing.

So, attention is all you need after all. You just need it again next quarter. And the quarter after that. And the quarter after that.