SaaS is being forced to grow up

The enterprises who bet on ungoverned AI just proved why systems + expert humans still win

Eighteen months ago, the narrative was everywhere. AI agents were going to make software platforms obsolete, the SaaS model was dying, and any company that had spent years and billions building enterprise software was about to get disrupted out of existence by a startup with a lightweight wrapper around an LLM. The term people used was “SaaSpocalypse,” and for a while, it rattled enough investors to send valuations swinging across the entire sector.

Then 25,000 customers showed up to ServiceNow’s annual conference last week, the biggest crowd in the event’s history. They were not there to eulogize the platform; they were there because they needed more structure, not less.

What ServiceNow’s executives described from that conference was not a world where AI had simplified enterprise operations; rather, it was the opposite.

A CIO had 900 AI pilots running across his organization and canceled every one of them, because nobody owned them and he couldn’t tell what any of them were actually doing.

And a CTO at a major bank built 30 production-grade AI agents and couldn’t put a single one into production because he couldn’t answer basic questions about what they had access to. ServiceNow’s president described it plainly as AI chaos, and the room full of analysts went quiet.

This is what happens when you adopt fast and govern never. And the response from those enterprises was to go looking for systems with guardrails, with accountability, with humans in the loop who could actually be responsible for outcomes. The most sophisticated technology buyers in the world, after an 18-month experiment with ungoverned AI, came to the same conclusion: they want defined systems and expert people involved.

I find this clarifying rather than surprising. Our leadership trio (at FOMO.ai) has seen many cases where businesses rely on data, but the data they actually have is appalling. So what happens? A data scientist translates the ask and builds a response that includes an explanation and caveats, which, in turn, should facilitate a human judgment call.

But data scientists have looked like easy targets for layoffs over the last 2 years, and companies have built AI agents to sit on top of that same data instead. Anyone can now ask & receive data analysis back… the problems, of course, are (1) they do not have the training or understanding of how the data might be misleading, and (2) they put more trust in a response from an AI than they do their former human colleagues.

The SMB version of the CIO with the 900-pilot problem looks different on the surface. There’s no CIO, no compliance department, no formal audit process. But there are three ChatGPT subscriptions across a team of five, an AI chatbot on the website that nobody has reviewed in months, social content going out that the founder hasn’t read, and an AI SEO platform generating reports that look impressive and that nobody fully understands. The tools are everywhere, and the ownership is nowhere, which is the same problem the enterprises had, just without anyone whose job it is to notice.

What those enterprises discovered, and what I think SMBs are starting to feel, even if they can’t name it yet, is that the value was never in the raw AI output; it was always in the judgment layered on top of it. I’ve written before about how Google’s Performance Max, left to run without human intelligence guiding it, optimizes toward the average across every dimension that matters. The same logic applies to any AI system operating without domain expertise sitting above it. It produces output that is technically functional and strategically mediocre, because it has no way to know the difference.

The enterprises tried to skip that layer and spent the last year cleaning up the results. The platforms they’re now paying to fix the problem, the governance layers, the control towers, the context engines built on proprietary data, are essentially just structured ways of putting expert humans back in charge of what the AI does. ServiceNow built an entire product around this, and the commercial response, according to their own executives, was faster than anything they expected.

For smaller businesses, the solution doesn’t require enterprise software; it just requires the same underlying principle: someone who knows the business, understands the customer, and can tell the difference between output that cleared the technical bar and output that’s actually effective. That gap is where most of the value in marketing lives, and it’s exactly where ungoverned AI falls short every time, regardless of the size of the organization running it.

SaaS isn’t dying; it is growing into an outcome-based system (which Sequoia just called the $Trillion company) rather than a point solution. The local attorney really doesn’t care about the underlying SaaS, and they certainly don’t want just another software subscription that they have to give headspace to; they just want sales & leads, and to know that is coming from a system they can trust and people they can hold accountable.

Dax is the Co-Founder & CEO @ FOMO.ai, and the author of 84Futures.com.

👉 Founder or operator? You can now schedule a 1:1 session with me.