In The Loop Episode 5 | The Rise Of Vertical AI Agents: Why SaaS Companies Should Be Worried

There's a fascinating shift happening in the world of software and technology, a shift so big that it challenges the very core of how businesses have operated for decades. At the heart of this transformation are vertical AI agents—the deep research agent launched by OpenAI is a perfect example of how this transformation will reshape everything.

Most businesses today rely on multiple different tools to run their organization and then hire many people to manage those tools. But what if a single AI agent could perform the function of all of those different tools and only take one person to manage? Most people today realize that AI agents might replace many of the things they've been used to throughout their entire lives, but very few understand how this is going to happen.

In the next 15 minutes. You are not going to be one of those people. This is In The Loop with Jack Houghton.

‍

‍

Why vertical AI agents threaten traditional SaaS

I've wanted to talk about this for some time because we all know that AI agents might eventually replace many of the things that people are used to, particularly technology companies, meaning SaaS (Software as a Service). That's because they typically provide a horizontal, general-purpose tool that can be used across different industries that supports a use case, like sending an email or organizing data.

This means that often, a business uses ten, twenty, or a hundred of these different tools all integrated and even hires functions of people to both manage those tools and all the data flowing between them. However, a single vertical AI agent could replace all of that software and much of the workload. This is what we call the verticalization of software.

The latest agents launched by Perplexity and OpenAI make this shift super obvious. For those who don’t know, the deep research agent (does exactly what it says on the tin) conducts lengthy reports with citations based on a massive amount of research. Now, this sounds relatively simple and similar to what we were used to, but if you go back to the old world versus the new world, there’s a clear difference.

In the old world, if you were given a problem and asked to do an extensive research report on that, you've got a researcher and a writer, and you spent days or even weeks going over Google or different sources trying to understand them. You took the relevant parts out of those sources, tracked every reference or citation, and maybe added all of that information into a spreadsheet to manage those citations. Then you wrote up a summary on a Google Doc, maybe even used an academic journaling paper, and then you emailed it to people for feedback.

However, deep research agents do all of those different workflows and replace all of those requirements for software through a single conversation.

The magic of how this works is that a person just thinks they're speaking to a single agent, but actually, it's an entire team of agents—a multi-agent system. Each agent has a very specialized purpose or job description.

There’s a boss called an orchestrator agent that essentially pulls in all these different agents to do different tasks at the right time. These agents also have tools—pieces of software to perform specific actions. In this case, it might be a web search tool.

When the deep research agent is given a single task, it then asks, "How am I going to execute this task?" That’s called chain of thought. It then reviews its process and can also iterate and improve once it finds more information. It might pull in specific agents like the research agent, who has access to that web search tool. When it gets information, it might pull in the report agent to start writing up the report. Each of these agents is like a specialized unit.

Perplexity does much the same thing.

‍

‍

Why Google’s business model is at risk from AI-driven search tools

If anybody's been using these agents, you'll probably notice that your workflow and key assets change quite quickly. This threatens Google and other legacy systems outside of that search context.

The user no longer needs to search through blue links and frustrating articles that don't get to the point. These AI research assistants integrate all the functions that you need—searching, reading, note-taking, summarizing, and citations—all into one super simple experience.

If you look at adoption statistics, they tell a very interesting tale. Gen AI tools have seen huge growth. ChatGPT alone has 400 million monthly active users, and a recent Google survey found that 93% of Gen Z and 80% of younger millennials ( age 17-30) are using at least two AI tools every single week. For companies like Google and other big players, this creates a classic innovator's dilemma.

Take Google, for example. They make most of their money through ads—specifically from people clicking on those blue links. But now, the only way for them to compete is through a different pricing model. The impact of tools like ChatGPT, Perplexity, and other emerging AI search tools comes down to one simple fact: user behavior has changed.

We now want to speak to a chatbot for searches instead of going through blue links. Initial research shows that when Google displays an AI-generated overview—those simple summaries you see when you search—people do not click those blue links. Specifically, click rates have dropped from 21% to 10%. In other words, users are just finding what they need directly in the AI summary, which obviously threatens Google’s entire business model.

In each domain, a vertical AI system replaces not just a single product but an entire stack of software. Let’s take an example.

In healthcare, doctors traditionally use multiple horizontal systems—patient medical record systems, EHR systems, reference databases, and decision support software. A vertical AI health agent could handle all of that. It could read patient files, suggest treatments, order treatments, and update records—all through a single conversation.

The same applies to law. An AI tool like Harvey could read all the contracts, highlight risky clauses, draft specific terminology and language—all in one seamless workflow. There’s no longer a need to rely on all those horizontal systems like Word, email, contract databases, or even manual reading with researchers, analysts, and employees.

At Mindset AI, we’re seeing this happen in real time because we work with software companies to help them launch AI agents into their products quickly.

And the big players can see this happening too. We’ve already seen classic defensive moves—like Microsoft launching Copilot features into Word, Excel, PowerPoint, and Outlook. Google’s doing the same. But if you’ve used those features, you’ll know they’re not compelling. They just don’t drive you to change your workflow.

As I’ve said a few times in different episodes, we’re now creating entirely new workflows and new key assets. We’re no longer trying to use the same old assets of the old world. I think it’s far more compelling to go all in with a conversational interface and start thinking through what those new assets will be.

I believe the future is a world with no interface at all—just telling an AI what needs to be done, or the AI generating the user interface on the fly.

‍

‍

A hybrid model for SaaS companies

Yet, in many ways, SaaS companies today are uniquely positioned to capitalize on AI. They often possess proprietary data sets and lots of deep organizational information—like Salesforce CRM data, IP, thought leadership, and policies.  This allows them to create domain-specific vertical AI agents that outsiders will struggle to match. Mindset Ai is really close to this problem. We work with SaaS providers to launch AI agents to their customers quickly inside their products.

I think one likely future is a hybrid model. Horizontal SaaS is going to be augmented by AI, copilots, and agents. So that familiar SaaS interface may remain, but an AI agent handles much of the behind-the-scenes complexity.

You could even describe the agents as intermediaries: running tasks within existing platforms. You might still open your product management software, but instead of clicking through lists and menus, you say “Prioritize and assign tasks for this sprint based on X, Y, Z factors,” and the AI will just handle it via APIs.

So in this world, the SaaS solution endures, but its role shifts from an everyday user tool to an AI-assisted backend.

However, for me, the most compelling alternative is that many horizontal SaaS offerings get completely unbundled by AI solutions—replaced by very specialized vertical AI solutions that just align with user needs better.

If an investment analyst could create rapport with an AI so they don't have to use a clunky BI system, Excel, Google, and all the other systems, I think they would. These new AI entrants can very easily leapfrog all of the software companies that exist today that have glued themselves into massive stacks with other providers by just doing the job of all of them.

‍

‍

Conclusion

It's conceivable to me that in just a few years, you might not need a CRM. A sales team will just have an AI agent that manages contacts, the personalized outreach schedule, follow-ups, updates, and records. And if a person needs to report sales numbers to their boss, their customers, or whoever, the agent might just generate a report and show you it on the fly.

I don't think Google or Salesforce or SAP, or other big tech players are over—but they're going to be facing a moment of reckoning. If software providers do not make those big bets today, they might find at the end of their contracts that many of their customers have already moved on to other options.

Anyway, that was this week's episode of In The Loop. I hope you enjoyed it and I’ll see you next week.