I had a very interesting conversation with a friend who is currently visiting from Seattle. For context, he's not a developer himself, but he's very much connected to that world. He runs a recruiting company, the kind that finds work for developers.
The conversation quickly turned to AI, because of course it did.
I wanted to know how it had impacted his work, his company, and what he thought the future was going to bring. He shared that the company had downsized in personnel, which of course wasn't shocking to hear, but that they had finally turned a profit. He sounded pleasantly surprised by it. The current AI tooling had allowed him and his employees to outperform the last few years with relative ease.
Of course, I couldn't just let the conversation stop there. I needed to know more.
It's no secret that AI usage is heavily subsidized. From the free models you can access on websites, to subscription plans that are vastly cheaper than paying for tokens as you go, it's pretty clear that we're living through a period where these companies are racing for adoption more than they're racing for profits.
One has to wonder what the long game looks like.
It's hard to imagine so many brilliant people building these companies without a vision for how this eventually becomes profitable. Just like the dotcom era, this one will eventually find its balance. What that landscape ends up looking like, and how we'll adjust to it, is anybody's educated guess.
My friend, however, falls on the side of what I would call a resurgence of the high cost of building software.
He believes, and he's not joking, that eventually what companies save in salaries by running much smaller teams, they'll end up paying back in AI tooling.
Some would call that future dystopian. Fewer people working, higher costs of living, and software becoming dependent on expensive AI subscriptions.
I happen to disagree with him.
Not because I don't think the subsidies will end. They simply have to.
I disagree because I think the genie is already out of the bottle.
A few days ago, Moonshot announced Kimi K3, an open source model that appears to rival Fable, and OpenAI's newest frontier model. Now, you may have your reservations about that claim. So do I. But after watching benchmark videos and the avalanche of tests currently making the rounds on YouTube, it doesn't really seem like Moonshot is selling smoke.
I say this knowing full well that having excellent open source AI models is only half the solution.
The other half is having the hardware to run them.
Frontier models aren't going to run on the old computer gathering dust in your closet. But who's to say groups of developers, entire communities if you will, can't join forces and co-own a little AI monster?
It doesn't seem far fetched at all.
Imagine if 
That doesn't sound unreasonable to me.
So I guess I'm both agreeing and disagreeing with my friend.
I do think the cheap AI era is going to end, and probably sooner rather than later.
But when the dust settles, I wouldn't be surprised if we find ourselves in a much better place anyway.
And hey...
We might even own our data this time.
MenO
