The AI Divide Isn't Just About Speed—It's About the Unsexy Problems Nobody Talks About

Field service scheduling is broken. Everyone knows it, but Jason Lemkin over at SaaStr recently put numbers to a bigger, stranger story: total software spend is accelerating to 15% growth, the fastest in a decade, while public SaaS stocks get obliterated. It's a paradox that makes sense the moment you stop looking at "software" as one thing. The market has bifurcated into two camps—one tapping AI budget and re-accelerating to wild numbers, the other running an 18-month-old playbook and waiting for a recovery that isn't coming.

Lemkin's breakdown of the 10 dynamics driving this split is essential reading for any founder. But it's also missing something crucial. The macro narrative—AI natives surging, pre-AI platforms catching a tailwind, legacy SaaS dying—is only half the picture. The other half is the thousands of unsolved, high-severity operational problems that still exist in the AI adoption lifecycle. And that's where the next wave of massive outcomes will come from.

We track over 22,698 problems across 94 industries, and the data tells a clear story: the pain isn't just about building a better CRM agent. It's about the unsexy stuff—AI hallucination in customer-facing roles, legacy SaaS integration hell, industry-specific workflows that generic agents can't touch. These problems have an average severity of 3.8/5 in Communication software alone, and they're not going away because someone vibe-coded a prettier leads tab.

Take AI inaccuracies in customer support. Lemkin talks about QB, SaaStr's AI VP of Customer Success, handling front-line support for 120 sponsors autonomously. It's impressive, but our data shows that "AI chatbot inaccuracies" is still a 4.2/5 severity problem across industries. The average team isn't SaaStr—they don't have 30 days to train an agent and iterate daily. They deploy a generic chatbot, it hallucinates once, and trust evaporates. The opportunity here isn't another chatbot. It's a tool that constantly audits and corrects AI outputs in real-time, or a vertical-specific agent that's pre-trained on industry jargon and compliance needs.

Which brings us to the biggest blind spot in the article: vertical AI agents. Lemkin mentions niche markets being 10x to 100x larger now because agents add so much value, but he doesn't go deep. Our data says the real gold is in hyper-specialized problems. "AI lease abstraction for commercial real estate" has a severity of 4.5/5. "Automated medical coding audits" sits at 4.3/5. These aren't problems you solve with a generic agent that books meetings. They require deep domain knowledge and integration with legacy systems that were never designed for APIs.

And speaking of legacy integration, let's talk about the "buy, don't build" advice. Lemkin is right that if a product works, you should buy it. But our data shows a surge in "no-code/low-code limitations for complex workflows" (severity 4.1/5) and "legacy SaaS integration with AI tools" (severity 4.0/5). Many teams are forced to build because off-the-shelf AI products can't handle their specific needs. The parking pass automation SaaStr built is a perfect example—no vendor would build that, so they built it themselves. The lesson isn't "never build." It's "build when the pain is so specific and unaddressed that no one else will."

This is where the next Twilios and Datadogs of the AI era will come from. Twilio became the communication layer for AI agents because it was already the API for messaging and voice. Datadog is the monitoring backbone for AI hyperscalers. Both solved unsexy infrastructure problems before they were cool. Right now, there's a similar gap in AI orchestration and reliability. We track 10,777 app ideas, and the ones in the top 5% are often about making AI agents work reliably in production—not just building another agent.

The article also frames the SaaS-pocalypse as a public market sentiment problem. And sure, software is trading at a discount to the S&P 500 for the first time ever, and stocks like Monday, HubSpot, and Atlassian got hammered. But our data challenges the idea that this means SaaS is dying. The severity of "software spend ROI measurement" is 4.7/5—customers are frustrated with waste, not with software itself. The demand for better tools is stronger than ever; it's just shifted away from mediocre point solutions and toward products that demonstrably improve efficiency or revenue.

Founders who want to be on the right side of this divide need to look beyond the AI spending frenzy and focus on the pain points that generic agents will never solve. That means talking to users in specific industries, not just vibe-coding another horizontal CRM. It means building API-first so agents can actually integrate (as Lemkin rightly emphasizes). And it means accepting that the daily grind of training and iterating on AI isn't a bug—it's the moat. Consistency beats brand every time, and anyone can copy consistent iteration. The ones who do it for problems that hurt the most will be the ones who re-accelerate next.