Beware of AI Marketing Traps

A Quick Side Tangent

I’m sure you’ve already read enough about how AI is going to take your job to ruin your life and eventually take over the world. It’s a very skeptical way to approach the situation but I understand the concern and reason for worry.

I tend to approach AI with a similar sentiment that Mark Andreessen has. He believes that the quality of life for civilization will be improved overall and we will have better quality jobs. Just like when the car was invented, or the printing press, or any other disruptive invention, jobs didn’t simply just disappear, they transformed into jobs that elevated the quality of the workers life.

Does this mean there won’t be growing pains, nope. Does this mean people will lose a ton of jobs, probably. It also means the overall quality of life for our society will be elevated, which from a holistic perspective is a good thing.

People getting laid off always sucks, I’ve been really close to that fire before and it’s the most stressful situation that seems unbearable. But humans are remarkable at “figuring it out” when they have to. This might sound cruel to say and I’m well aware there’s a lot to unpack in that sentence. It’s not easier said than done and it doesn’t invalidate the feelings of dread that one goes through after being laid off. But I believe in humanity and I believe that people are able to get through hard time.

What I’m getting at is I think AI will have a huge positive effect on society, but there will be growing pains.

I Built My First AI App

Before I dive into how marketing teams trick you into thinking something is AI, I’m going to give some background.

The Idea

A couple of months ago I had a cool idea for an app that I thought would be valuable for students. I graduated from BYU about 2 years ago and I recalled the pain of having to comb through all my notes to memorize them. I was always conflicted whether I should study my notes, the professors power point slides, the reading material, or any other resource I was given.

I remembered that it was really hard to challenge myself on the material if no practice tests were provided. I pretty much resorted to memorizing all my notes. This worked out pretty well to get the grade, but 1) it wasn’t fun, and 2) it didn’t really facilitate long-term growth and learning

So I came up with app idea that turned into Quizzzard. This is an AI app where you can upload your notes from class and quizzes will be auto-generated for you based on your notes. This way you can test yourself on your own material to better prepare you for exams.

Disclaimer

I found out their is another company called Quizard AI… So I’ll have to figure that out. I’ll probably just buy a new domain and change the name of my app since what my app does is a little different than Quizard AI. My apps also just a small hobby for now, could it grow into a huge business, maybe. But I wanna steer clear of any conflict for now.

The Process

I started off building Quizzzard. I spun up a simple Next.js project and built out a basic frontend with a simple file upload form to submit.

Then I started building out the API endpoint to process a file and send it to an LLM with a nifty prompt to generate a quiz for me. I stored all the file uploads on an AWS S3 bucket before sending them to the LLM

I tested a bunch of different LLM, mainly Anthropic’s Claude Sonnet, xAI’s Grok, and Google’s Gemini. I really wanted to focus on the user experience and make sure they had a wide range of different file types they could upload.

This was the tricky part since Claude is REALLY good a process unstructured data like PDF’s but cannot ingest images. Grok can ingest images though, so I started with using Claude for PDF files and Grok for png, jpeg, and other image files. I also allowed for Word Doc file uploads, but I just converted those to PDF’s and sent to them to Claude on the backend.

And viola!! A quiz generated from quiz notes!! Pretty basic, but a good prompt goes a long way!

This is still very much an MVP (minimum viable product) but something good enough for people to test out!

What does this after to do with Marketing Companies talking about AI

So what does this story have to do with the actual title of the article. Well, Quizzzard is a cool idea that uses AI… but it is NOT AGENTIC AI!!!!

Quizzzard is a basically a predefined process in which I consistently feed an LLM data with a prompt and get a result. It’s predictable every single time. It’s basically an automation that integrates with an LLM.

Many companies will claim to incorporate some revolutionary AI features, but it’s really just an automation that integrates with an LLM. There’s a BIG difference between this and Agentic AI.

Agentic AI is able to act autonomously to think and take actions to achieve a specific goal. My goal with Quizzzard is to create an Agentic AI eventually, but for now it’s really just an LLM automation. Still valuable and a cool MVP, but definitely not agentic.

There are a ton of companies that are building cool things and are building agentic AI features that are really cool. But some companies try to pass off LLM automations as agentic AI and it’s just something to be aware of and watch out for.

How to Tell the Difference

Obviously you don’t have access to a companies code base so it’s sometimes hard to know if their product is truly agentic or an automation.

Here are a couple of tips to help you know the difference:

1) Ask the right questions

If you’re being pitched on a new AI product, ask questions to make sure you understand the capabilities of the AI feature. It’s important to know if the feature can perform independent actions to achieve a goal without you hand-holding it.

Some examples are…

• “How will I have to maintain <feature> to make sure it operates correctly?”
• “What’s the strategic goal that is outlined for <feature> and how will it benefit my business?”
• “Can I customize the outcome of the <feature> to fit my need?”

2) Test the product

You can usually tell if a feature is agentic just by testing out the product to see how complex it is. If it doesn’t take a lot to set up on your end, it probably isn’t agentic.

3) Talk to other people that have used the product for a while

A great way to get a second opinion is to ask for referrals to talk with other businesses that have used the product for a while. They have more experience with the product and can get you a better understanding if the product is truly agentic or not.

Conclusion

I know I tend to go on side tangents when I write, but I got to the point eventually right? Just always watch out for marketing teams at companies trying to pitch you Agentic AI. Their product very well could be, but it’s important to validate that then just take them at their word. As they say… Trust but Verify.

Also, just because a feature isn’t agentic does NOT means it’s not valuable. There are a lot of cool product features out their involving LLMs that are really cool and are not agentic. Just don’t let the sales guy pass it off as agentic when it’s not. The main focus should always be the impact it provides your business, so if there is a product that is not agentic but still delivers impact, it is probably still a good investment.