My good friend works at a mid-market SaaS company. Vertical software, sold to a specific type of institution. Think help desk, asset tracking, workflow management. Bread and butter stuff that keeps operations running.
Not long ago, his company announced their multi-year roadmap to become fully AI-enabled. The whole company got the presentation. Phases. Milestones stretching to the end of the decade.
Completely cooked, I told him.
The product is a mess. Dozens of microservices on Java with a React frontend, the whole nine yards of enterprise architecture astronautics. Dozens of engineers maintaining what is essentially a CRUD app with some workflow logic on top. Yes, there are still shops running top-to-bottom legacy stacks with armies of developers in 2026. Lots of them. And they employ huge teams to keep the lights on because the architecture is so fragmented that nobody can reason about the whole system anymore. The complexity is self-perpetuating. You need more people to manage the complexity that was created by having too many people.
Did I mention their big AI initiative so far? A chat assistant built into the product. Welcome to 2024. While they're shipping what every SaaS company was demoing two years ago, agents worldwide are learning to do the actual job of the people who use their software.
A multi-year plan to adopt AI in this kind of organization is like announcing a multi-year plan to launch a website in 2005. By the time you're done, the game is already over.
Nobody at these companies sees it because they're too close to it. Take any vertical SaaS product and strip away the frontend. Remove the dashboards, the forms, the navigation, all of it. What's left? A database. Workflows. Business rules.
That's it. A system of record for a specific domain with some logic about how work moves through the system. That's the actual value. Everything else is wrapping paper.
It's worth thinking about who actually uses that software today. An administrator or a small team of them. They log in, they click around, they process tickets, they update records, they run reports. Maybe some people in the field clicking around a mobile app. All day long, every day. AI agents don't need any of that UI. An agent can talk directly to the database. It can execute the workflows. It can apply the business rules. It can do what that administrator does, except faster, around the clock, and without ever losing context.
What do you actually need to replace one of these products? It's shockingly simple. The data model (which is just the domain knowledge of how that industry operates), the workflow logic, and an agent layer on top. One competent engineer who understands the domain could build a working prototype over a long weekend. I'm not exaggerating. The thing that took 20 C# microservices and 40 engineers to build can be a single Ruby on Rails app with an MCP interface. Not that the original was ever that complex. It just got built that way because that's what happens when you let enterprise architecture culture run unchecked for a decade.
I told my friend exactly this. And then I told him what I'd actually do if I were in his shoes. Don't quit. Actually, one sec, that's bullshit. First I told him to get the hell out of there as soon as possible. But you know, the labor market seems to be kind of shit unless you're completely junior or an incredibly based veteran like me.
So maybe don't quit. But don't bitch and moan about the three-year roadmap on your company's Slack, either. Instead, run your own private skunkworks. Take everything you know about the domain, all the tribal knowledge you've accumulated, and build the replacement yourself. Not a clone of the existing product. Something fundamentally different. An AI-native version that replaces the human administrator, or at least turns a five-person job into a one-person job.
Crank it out on nights and weekends. Ralph loop and swarm that shit. Make it look polished with Superdesign. Get it actually working with real data.
Once it's ready, present it internally. Yep, show it to your bosses. Why? Because your company already has the distribution channels. They already have the customer relationships. They already have the sales team. What they don't have is someone who actually understands that the product needs to be rebuilt from the ground up, not given a three-year course of AI sprinkles while 40 engineers keep the existing Rube Goldberg machine running.
One thing I guarantee: when leadership sees a working demo and asks "how long did this take you?" and you say "two weeks," that lands differently than any strategy deck ever could. Especially when their current product took five years and a hundred million dollars to get where it is.
Keep this in mind: you win either way. Best case, you get recognized, you get promoted, maybe you end up owning that new product line. I mean, you did just create massive value using assets the company already had, right? Worst case scenario, the company is too slow or too political to act on your initiative. Fine. Throw that shit away. Cause regardless you just built up skills that are invaluable to any forward-thinking employer, or that can be applied to launching your own startup. The distribution problem is real, but it's solvable. (Just poach the top sales guy at your current company, ha!) The product problem is the hard part and you solved it already.
Either way, sitting around waiting for a multi-year AI roadmap to save you is the only way you lose.
Don't take my word for it. The SaaS industry just shed $2 trillion in market cap in a single month. Investors are waking up to the fact that most of the value in these companies is locked in the customer relationships, not the software itself. The software is the part that gets replaced. If you're an engineer at one of these companies and you can see the writing on the wall, stop waiting for permission. Build the future yourself and let the results speak for themselves.