And it is a fact that in any IPO S-1 prospectus, there will be a section that lists all kinds of risk factors for why its stock may not be a good buy. Most of these are boilerplate warnings against financial performance or acts of war and nature, but some apply more to the company itself.

We are seeing the possible birth of a new boilerplate warning: while large language models (LLMs) that can help businesses grow might indeed grow businesses, they also might hurt them.
ServiceTitan, a cloud service startup whose IPO documents with the SEC landed on Monday, had a 1,150-word risk factor on how the use of AI specifically with generative AI could adversely affect its business. It warned that LLM hallucinatory behavior could produce "inaccurate" information and engage in "discriminatory" behaviors; its LLMs could infringe on others' copyright or intellectual property; and using LLMs means exposing more data to potential hacks and harm. Then again, if it can't get enough data, it may not be able to continue offering AI products or building new ones. Furthermore, the company itself warns of an eventual possibility that its employees or contractors might inadvertently expose its customers' private information with third-party systems that could then lead to security breaches, or use that data for their model training. ServiceTitan cautions its use of AI may one day be an affront to social and ethical issues. Or that future rules will ultimately cost it money.

ServiceTitan also has a concern that it might not be able to hire the AI experts it needs for its products, and even if it could, that talent will be expensive. ServiceTitan is advising its AI products rely on third-party services it doesn't control — and specifically names Microsoft and OpenAI. So, if they become unavailable or have issues, that's a risk too.

These are fascinating warnings because ServiceTitan sits at the heart of the industries that GenAI, specifically LLM-driven AI agents, will be predicted to conquer. ServiceTitan provides software for businesses in the field service industries, and for the most part, these are small businesses: think construction contractors, HVAC professionals, landscapers, and the like. Its software does everything from marketing and managing customer relationships to helping with support and accounting. (ServiceTitan declined further comment.)

AI agents are now marching so furiously towards all things sales, marketing, and CRM that the king of the CRM market, Salesforce CEO Marc Benioff, said to TechCrunch that his company alone could have a billion AI agents working at Salesforce's customers' companies in a year.

ServiceTitan has been providing AI-based services since at least 2023, under the brand name Titan Intelligence. In October, it also released a suite of AI agents for sales, customer service, and call centers.

But LLMs were designed to study words, images, and data and create their own versions of things. That is, they were designed to make stuff up. OpenAI’s scientists are busy racing toward general intelligence — mimicking humans’ ability to infer even better — meaning they are working on AI that they hope will be even more creative.

Yes, LLM reliability for use in business is something that will be worked out over time as more companies move to specialize in agents in which reliability, not creativity, will be paramount. But note, too, that those selling AI bots today are really telling the quiet part loud: At this early stage, AI adoption creates as many problems as it solves.