For years now we've been hearing about how SaaS is dying in part because OpenAI's just gonna build everything. Project management, coding tools, sales platforms, etc. And for the last year and a half or so, Anthropic's been added to this argument as well.
This is the consolidation thesis: the idea that the big labs, who know LLMs better than anyone, have more money than most, and have attracted the top talent in the industry, are just able to build anything they want faster and better than you. And they might not be doing it now, but if it becomes clear that there's money to be made, they will.
In some ways, this argument has always existed against startups. "Can't Google just build this?", investors would ask. But I do warrant that it's a bit different now and the argument feels more solid.
But I'm not here to debate whether the big labs can or can't compete with SaaS platforms, nor to argue that SaaS is in fact not dying (or dramatically changing). I am here to say that the consolidation thesis seems to be going down the drain.
Why is that, you ask?
Well, consider how up until around the end of last year, Opus was it. If you wanted to get the best quality when coding with LLMs, you needed to be running Opus. Then Codex five-point-whatever-it-was came along and some people were saying it was a lot better than Opus and made the switch. But then the whole Pentagon thing happened and for like a week a lot of people cancelled their OpenAI subscriptions (do we even remember this?) and it felt like Anthropic was ahead again.
And now, April has come around and apparently Claude Code sucks, Anthropic hates open source, they hate developers, and even their customers. If you want good results, run Codex, they say.
Which leads me to my point: are we going to want to be stuck with one model provider when the other one might just leapfrog it tomorrow?
Not only that, these labs are in the business of making models and selling inference, so if the vertical product you're using from them no longer makes financial sense as a side quest, they will kill it (hey Sora!).
Developers have proved that a) it's easy to switch between model providers and b) we seem to regularly want to do so. I don't see how it would be any different for other products.
It does seem like there's space for a layer above the model providers, solving specific pains, and leveraging whatever model provider is best at that point in time, either in the background or giving the user a choice. This is already happening, but some would argue it's just a transitional state. I'm arguing this layer has some merit for the longer term.
Now, as for the SaaS market as we've seen in the last decade, who knows? It's definitely changing. There probably will be some consolidation, and there will be more tailor-made software built in-house.
So I don't know if SaaS is dying or not, but if it is, I don't think it's because OpenAI and Anthropic will eat your lunch, at least.