With the exception of a short time early on, I’ve always been a solo founder.

I decided to go it alone after my original cofounder and I had an ideological disagreement we couldn’t reconcile—he wanted to focus on consulting, and I wanted to build forms. We quickly went our separate ways.

In the many years since our split, I have had zero regrets about the decision to build my business sans partner. I’ve had the freedom to chart my own course, make my own decisions and yes, own my mistakes. I’ve learned to embrace occasional loneliness and work through the uncertainty of what to do next.

But just because I’m solo doesn’t mean I’ve been alone on my journey. Quite the opposite. I’ve always sought insights from the people around me, whether they’re peers, mentors or others in the tech community whose opinions I respect.

Now, I have another tool in my arsenal: ChatGPT. By the time LLMs entered the scene a few years ago, I’d already been running Jotform for close to two decades. I knew I had the mettle to lead my company by myself. Even so, I've come to think of ChatGPT as the business partner I never had—one that I never have to worry about annoying, or vice versa. Here’s how I use it.

How We’re Using LLMs (And Why We’re Wrong)

Truthfully, most people aren’t using ChatGPT to its full potential. Most are using it as a glorified search engine. I’ve come to think of it as a vending machine—people punch in a request (“write me a marketing plan,” “summarize this document.”) and wait for the answer to pop out. If the output is mediocre, they shrug, then immediately hop on LinkedIn to deride AI as overrated.

But LLMs are only as good as the direction they’re given. New research from KPMG and the University of Texas at Austin found that while AI adoption is high—90 percent of the 2,500 employees surveyed—only 5 percent qualified as what the researchers deemed “highly sophisticated users.”

Sophisticated use, in this case, includes those who write more involved prompts, have clear objectives and use AI as a reasoning partner rather than a shortcut. “In simple terms, it’s less about the tool itself and more about how professionals think and make decisions with it,” the authors write.

Today’s LLMs are smart, and they’re only getting smarter. According to researchers at METR, an AI evaluation organization, the complexity of tasks that leading models can complete independently has been doubling roughly every seven months. None of that, however, will substitute for having a capable human guiding it.

So if most of us are using AI wrong, what should we be doing instead?

One of the main issues with the “search engine” approach is that it doesn’t provide the context AI needs to be truly helpful. Instead of thinking of ChatGPT as Google in a different outfit, think of it more like a new employee—smart and capable, but without the information required to do its job well.

Providing context is crucial to maximizing the potential of AI. In fact, that’s the reason we added robust training options to our AI product, Jotform AI Agents . Users can manually enter any knowledge about their business or service, upload related documentation files, link to a live web page, or even add a Q&A section to address the most frequently asked questions directly. After you give context to the AI agent, you can sit back and watch your agent handle all the customer support autonomously.

A key finding of the KPMG/UT Austin research was that top AI users tend to have longer interactions with LLMs—they write more involved prompts, and are intentional about switching to a different model or tool depending on the task at hand. Whereas less sophisticated users may enter a poorly-informed prompt (and quit when the answer isn’t up to snuff), more effective users tend to regard LLMs as a thought partner who needs a little guidance.

For example, say I’m debating releasing another product. If I prompt an LLM with something vague—”Should I build another product?”—I am sure to get something lackluster in return.

Compare that with this: "I run a bootstrapped SaaS company with one core product that's mature and profitable. We have a large existing user base and a few adjacent problems we could plausibly solve. I'm weighing whether to expand or stay focused. What are the risks founders typically underestimate here?" You can probably guess which one will yield the higher quality response.

Using ChatGPT As A Partner

Another sign that you may not be using ChatGPT to its full potential? You don’t push back on its answers.

OpenAI recently retired its most sycophantic model, GPT-4o—while its creators said customers were drawn to the LLM’s “conversational style and warmth,” a less generous description would be that GPT-4o was an overly-flattering “yes man,” regularly falling all over itself to validate users’ correctness. Frankly, the newer updates represent an improvement. Being constantly told you’re right about everything isn’t going to help you grow your business.

That’s because one of ChatGPT’s most helpful roles is that of a sparring partner, offering the chance to test assumptions and play devil’s advocate. Some of my most useful interactions with ChatGPT have started with something like: "Here's what I'm planning to do. Make the strongest possible case against it."

Skilled AI users aren’t passively accepting the model’s answers, but actively working to shape its thinking, the KPMG/UT Austin authors explain. By showing it examples of stellar output and establishing clear parameters around your queries, you’re creating an ideal environment to help AI help you. “The value came from guiding the model over time, not from asking ‘better’ one-off questions,” they write.

Any psychologist will advise against assuming people can read your mind. The same goes for AI. It’s smart, but it’s not omniscient, and it doesn’t know what you want unless you tell it. To get the most out of this powerful new tool, stop using it as a search engine and start using it as the partner it’s capable of being.