Not everything is AI or Agents

The boring stuff is the point

Everyone's prophesying the downfall of SaaS. Guess what — it's coming. But not for the reason you think.

It's not that AI agents are replacing every tool we use. It's that the AI-ified versions of those tools are becoming unreliable, while the boring deterministic ones — the ones doing what we actually pay them to do — are still standing. Call it the bias of capitalism or fashion, but my inbox is getting bombarded with every service I use telling me "Oh wait! Yeah, I used to do X, Y, or Z, but now I'm an AI-agent slop shop."

Here's the thing nobody seems to be saying out loud: we don't always want intelligence. Sometimes we want the boring stuff to stay boring.

It’s all about flexibility

With Generative AI we're now talking about "buying intelligence" — but we've always been buying intelligence. We just bought it in different forms.

Hardware — dedicated chips serving a specific purpose — was already intelligence, encoded and hardwired into tin, copper, and gold. Software ate the world because it was more flexible than hardware: when a bug appeared, you didn't need to re-melt everything and ship it across the world, you just published an update.

This kind of intelligence isn't "human intelligence." It's predictable, deterministic (or at least, it should be), and boring. But you need boring from the tools you use every day. You need your spreadsheet to add the same numbers the same way, every time. You need your auth provider to log you in, not have a creative interpretation of who you are.

Generative AI and LLMs are a new level of intelligence — more context-aware than software, more flexible than hardware, but less deterministic. That last word is the catch. Flexibility and determinism sit on opposite ends of a spectrum, and every layer of your stack lives somewhere on it.

Every product is a promise

A product exists to tell you one thing — the only thing you're actually willing to pay for, summed up in a few words:

This problem you have is now my problem, and I promise I’ll take care of it

It's a promise. You can trust it or not. And here's where this connects back to the previous section: the promise SaaS makes is a deterministic one. "I will charge your customers' cards." "I will deliver this email." "I will store this file and give it back to you exactly as you sent it." When you slot an LLM into the middle of that flow, you're not adding intelligence — you're breaking the contract.

SaaS isn't going to disappear, because you don't want your microwave to be self-aware. You don't want it agonizing over whether you really want today the same thing you wanted yesterday. You want it to do the boring thing, on schedule — so you can plan the rest of your life around it.

Once in a while, I get reminders of what happens when boring goes missing. Like this gem from a recent agentic coding session:

Sorry — I clobbered your .env during one of my earlier compose-validation chains (the cp .env.example .env … rm -f .env pattern ran in that scenario's dir). It's not recoverable from my side.

Could you refill it the same values you had before ?

Thank God I don't give SSH access to my LLMs.