Vertical AI Agents Could Be 10X Bigger Than SaaS

The Lightcone hosts recently discussed the potential of vertical AI agents, a new business model emerging as AI models rapidly improve. They explored how these agents might impact existing SaaS companies, identified suitable use cases, and even suggested that this category alone could produce 300 billion-dollar companies. It's a big idea that could change a lot of things.

The Rise of Vertical AI Agents

Things have been changing fast in the AI world. Every few months, it seems like things get better and better. Now, we're talking about full-on vertical AI agents that could replace entire teams and functions in big companies. This progress is pretty amazing to think about.

Foundation models, which are the base AI systems, are starting to compete more directly. For a while, OpenAI was pretty much the only big player. But now, we're seeing more competition, which is good for everyone. More competition means more choices for users and more chances for new companies to start up.

Jared, one of the hosts, is really excited about vertical AI. He thinks many people, especially new startup founders, don't fully get how big vertical AI agents are going to be. It's not a brand new idea, and some people are already talking about it, but the world hasn't quite caught on to its full potential. He believes there will be 300 companies worth over a billion dollars just in this one area.

Lessons From SaaS

To understand how big vertical AI could get, it helps to look at the history of SaaS (Software as a Service). Many people don't realize how huge SaaS is because, as consumers, we don't use many SaaS tools. They're mostly built for businesses. For the last 20 years, Silicon Valley has mostly been funding SaaS companies. Over 40% of all venture capital dollars in that time went to SaaS, and it created over 300 SaaS unicorns (companies worth over a billion dollars). That's way more than any other category.

The big change that kicked off the SaaS boom was a technology called XML HTTP Request, or Ajax, in 2004. This allowed web browsers to build rich internet applications that felt more like desktop programs. Before that, web apps were slow and clunky. This new tech led to things like Google Maps and Gmail and really set the stage for SaaS.

Paul Graham, a well-known figure in the startup world, was one of the first to see the potential of web apps. He even created one of the first SaaS apps back in 1995, but it wasn't great because the technology wasn't ready yet. It wasn't until 2005, when XML HTTP Request became widespread, that SaaS really took off.

This situation with large language models (LLMs) feels very similar. LLMs are a new way of computing that lets us do things that were fundamentally different before. Back in 2005, when cloud and mobile computing became big, people wondered where the value would go and what opportunities there would be for startups.

Looking back, billion-dollar companies created during that time fell into three main groups:

• Obvious Mass Consumer Products: These were things like documents, photos, email, and chat that used to be on desktops but could clearly move to the web and mobile. Interestingly, no startups won in these areas. Big companies like Google, Facebook, and Amazon took over all of them. For example, many companies tried to bring Microsoft Office online, but Google won.

• Non-Obvious Mass Consumer Ideas: These were ideas nobody predicted, like Uber, Instacart, DoorDash, Coinbase, and Airbnb. These came out of nowhere, and big companies didn't even try to compete until it was too late, allowing startups to win.

• B2B SaaS Companies: This is the biggest group, with 300 billion-dollar companies. There isn't one big "Microsoft of SaaS" that does everything for every industry. For some reason, these companies tend to be different from each other, which is why there are so many.

Salesforce is often seen as the first true SaaS company. In the early days, people didn't believe you could build complex business applications over the cloud. It was a new idea, and the early web apps weren't great. But Salesforce proved it could be done, opening the door for many other vertical SaaS solutions.

Why Incumbents Didn't Win in B2B SaaS

One reason big companies didn't dominate B2B SaaS is that it's just too hard to do so many different things as one company. Each B2B SaaS company needs people who are really good at one specific area and care a lot about small details. For example, why didn't Google build a competitor to Gusto (a payroll company)? Because no one at Google really understands payroll and all its complex rules. It's just not worth it for them; they'd rather focus on a few very large categories.

Also, older enterprise software like Oracle or SAP tried to do everything, which often made them hard to use. They were a "jack of all trades, master of none." This created an opportunity for B2B SaaS companies to build much better, more user-friendly products for specific needs.

The Future of Work and Vertical AI

In the past, as a company grew its revenue, it usually had to hire more and more people. Many unicorn companies today, even with hundreds of millions in revenue, have hundreds or thousands of employees. But this might be changing with LLMs.

Instead of hiring many people for roles like customer success or sales, companies might hire more skilled software engineers who understand LLMs. These engineers can automate specific tasks that slow down growth. This could mean that companies will need fewer people to reach high revenue levels. We might even see unicorn companies run by only 10 employees, with LLMs handling many tasks.

This trend was already starting before LLMs. For example, one company used an engineer to handle marketing, and they were able to spend a lot on campaigns with a small team. This showed that smart engineers can find ways to get a lot done with less. LLMs can take this even further.

The Case for 300 Vertical AI Unicorns

It's possible that for every SaaS unicorn out there, there could be a vertical AI unicorn that does the same thing, but better. Most SaaS unicorns replaced older "box software" companies. Now, vertical AI agents could disrupt SaaS companies in the same way.

SaaS companies build software that a group of people use. The vertical AI equivalent will be the software plus the people, all in one product. This means that vertical AI agents could be 10 times bigger than the SaaS companies they replace because they will also take over a lot of the work that humans currently do.

Companies spend much more on employees than on software. So, these new, more efficient AI companies will need far fewer humans for things like data entry or approvals. This could lead to smaller, more efficient companies.

Current Examples of Vertical AI Agents

We're already seeing examples of vertical AI agents in action:

• QA Testing: Companies like Metic are replacing entire QA teams with AI. In the past, QA-as-a-service companies tried to make QA teams more efficient, but they couldn't fully replace them. Now, AI can, which means companies don't need to build a QA team at all.

• Recruiting: AI can now handle the full recruiting process, from technical screening to initial recruiter calls. This means companies might not need to build large recruiting teams in the same way they used to.

• Developer Support: Companies like cap.ai build chatbots that answer complex technical questions, reducing the need for large developer support teams. They learn from documentation, videos, and chat history to give good answers.

• Customer Support: While many companies claim to offer AI customer support, only a few can truly replace a human team. These advanced AI agents can handle complex workflows and are gaining traction in large companies, especially in specific industries like instant delivery marketplaces.

These examples show that customers often need very specific, tailored solutions. It's hard to build one AI solution that works for everyone, which is why we'll likely see many specialized vertical AI agents rather than one giant "meta-AI" company.

AI Voice Calling

AI voice calling is another area that's growing fast. Companies like Salient use AI to automate tasks like debt collection in the auto lending space. This kind of job, which is often repetitive and boring for humans, is perfect for AI. Salient's AI can make very accurate calls and is already working with big banks.

Just a few months ago, AI voices weren't realistic enough, and there were delays. But now, AI voice apps can meaningfully replace humans in many calling situations. This rapid progress is amazing, considering how new this technology is.

Finding the Right Vertical

For founders looking to start a vertical AI agent company, the key is to find boring, repetitive administrative work. If you can find a task like that, there's a good chance it could be the basis for a billion-dollar AI agent startup.

It also helps to have some personal experience or connection to the problem. For example, one company is building an AI agent to bid on government contracts because a friend's full-time job was to manually search for new proposals. Another company is building an AI agent for medical billing in dental clinics because one of the founder's mothers is a dentist and he saw how boring that task was.

Just like in robotics, where profitable robots often do "dirty and dangerous" jobs, for vertical SaaS, look for "boring, butter-passing jobs." These are the tasks that AI can take over and make much more efficient.