How I Feel About AI

I’ve worked at the forefront of an AI-first company. Our entire product revolved around using AI to improve customer service, and the company’s success was directly tied to how well the AI performed. You’d think that in a setup like this, AI would dominate the workload, but in reality, the AI portion of the job rarely made up more than 10% of the effort.

When you’re building something like a chatbot, it’s crucial to get the AI part right. After all, that’s what makes it feel intelligent to the user. But here’s the thing. Getting the AI right is just one piece of the puzzle. What’s even more important is ensuring the application actually works as expected. A chatbot that understands your request but can’t pull the correct information or complete the task is just frustrating.

It’s worth remembering that natural language processing (NLP) isn’t new. The technology has been around for decades, long before the recent surge of large language models (LLMs) or the flashy promises of generative AI. While generative AI is undeniably impressive, I struggle to see how it could have meaningfully improved the product we were building by simply generating more content.

AI is, at its core, a tool. A means to solve a problem. If you can’t identify a clear problem, you’re essentially building a solution in search of one. Every time I meet with a founder of an AI startup, I’m reminded of this quote:

“While none of the work we do is very important, it is important that we do a great deal of it.” – General Peckem, Catch-22

That line feels all too familiar in the world of AI startups. I’ve seen countless companies slap the label “AI-powered” on their products, not because it’s necessary but because it generates buzz. For example, I’ve come across tools that take user input and convert it into a SQL query. Sure, it sounds fancy, but it’s nothing a competent programmer couldn’t have built without AI. The end result? The company now needs both the programmer and the AI service to accomplish the same task—an expensive and redundant setup.

AI is at its best when it’s applied to real, existing problems. It can make your life easier by helping you write emails faster, generating boilerplate code, or breaking through writer’s block. These are practical, tangible benefits.

On the other hand, AI as a solution looking for a problem is just a flashy, overpriced distraction. It might grab attention in the short term, but without solving something meaningful, it’s hard to justify its existence.

At the end of the day, AI should always be viewed as a tool, not the end goal. If you have a clear problem to solve, AI can be transformative. But if all you’re doing is using AI to generate hype, you’re likely wasting resources on something that could have been done more efficiently, and effectively, without it.

---

Side note: I am writing a book on how effective chatbots work. You can follow the progress along on github.

---