It’s not just tech-focused companies — every business and brand will eventually have to grapple with if and how they will add AI to their website.
I sat down with our technical director, Vlad Moraru, to chat about how businesses should be thinking about implementing AI. Below, he shares his thoughts on what businesses owners and marketers should ask when considering AI features for your website, the limitations you should be aware of, and the state of AI in web development and what the future holds.
What Should Business Owners Know about AI?
Rachel: Let’s say we’re talking with a brand or a business owner who wants to add AI to their website. What are the things going through your mind? What questions do you feel brands need to be asking when they’re considering implementing AI in some fashion to their site?
Vlad: I think the first question would be: What problem do they want AI to solve, or what do they think it is going to solve? That is the first thing I would ask a client, and we would take it from there. I think AI should be treated in the same way you would hire any employee. Does hiring for the position make sense? Do we need the role? Where can we put the role in our company, and what is it going to do?
From there, I think they should find repetitive work tasks within their business—low-level tasks that can be handled by AI without human intervention—and decide whether the investment of implementing AI to take that task over is worth the savings in efficiency, labor, etc.
AI should be treated in the same way you would hire any employee. Does hiring for the position make sense? Do we need the role? Where can we put the role in our company, and what is it going to do?
Beyond the AI Hype: Purpose vs. Luxury
Rachel: Often, when people say, “Let’s add AI,” what they’re envisioning is a chatbot that maybe replaces a live customer service rep. Is this a common use case, or are there other features brands should consider so they don’t fall behind the competition? It feels like AI features are a bit all over the place right now.
Vlad: It is all over the place right now, because everyone wants the new shiny toy. I see it as being similar to when the internet was first created; physical shops compared to web shops were losing sales. There’s a fear of falling behind, of not acting fast enough to replace one source of revenue with another.
But again, you need a clear purpose. You need to know what this AI feature is going to do for you and what it’s going to help you with. Go back to the basics: first, you need to discover the issues in your company or website and find the repetitive tasks or problems that AI can fix for you. For example, a chatbot can fix the issue of a user searching through documentation. Instead of losing time searching, they can just ask.
However, it is not magic. You need to prepare the chatbot, give it the knowledge and documentation, and that needs to be kept up to date. AI can help with searching facts and being available 24 hours a day, which is good for business, but it cannot compare with a human sales agent yet. A good salesperson can sweet-talk you or can spontaneously offer a discount based on how the conversation is going; AI is not going to do that, at least not yet. Or, it needs to be trained to be a good salesperson, but is the cost of training worth it?!
You need a clear purpose. You need to know what this AI feature is going to do for you and what it’s going to help you with.
The “Magic” of AI
Rachel: Do you feel like most people view implementing AI as an intentional, heavy lift, or do they see it as “magic”? And as a technical director, what are you seeing that tells you AI isn’t ready to take over your job just yet?
Vlad: Everyone thinks it’s magic, especially when you start using it, because it is crazy what it can do. When I first started using it for coding, I was a bit scared, but after testing it, I realized it is not going to take my job at the moment.
I still see it like it’s running on a script. It pieces LEGO blocks together, but that doesn’t mean it is always making the best decision. It doesn’t see the big picture and still needs human direction. It can pull up best practices for a piece of code, but without looking at the bigger context, that piece of code might work great on its own but break the entire application.
It is a tool to make my life easier. Instead of taking a week to test something, I can do it in a day or two. But for a final product, it’s not there yet. I hope it won’t be 100% reliable until the next generation! [laughs] Right now, these are probabilistic models; I would say they are only 50% to 60% reliable.
[AI] can pull up best practices for a piece of code, but without looking at the bigger context, that piece of code might work great on its own but break the entire application.
The Future: Monitoring Agents
Rachel: I’ve caught myself comparing AI to a junior-level employee whose work I have to check constantly. Is the future just one person monitoring 20 different agents that write the code for us?
Vlad: I think that is going to be the future — monitoring agents doing the work and steering them on what needs to be done. The big decisions will still be made by people.
Near-future work will involve developers supervising these models. There are so many programming languages; a developer might learn two and migrate to another, but they won’t know every in and out. A machine has access to all of it instantly.
Currently, I use AI for code reviews and small, isolated tasks. When I tell it to do something small, it has 80% to 90% reliability. But if you tell it to do something that affects five other things, it might break something or even change code without telling you.
The main issue is memory and context. When AI starts working on something, it can forget the context it had five minutes ago and start messing up the task. They are trying to mitigate that with bigger “context windows,” but there are still limits. When a limit is reached, they have to compress the session, and it loses a lot of context to start a new one and continue. It isn’t “Skynet” yet; it’s still following a script, but it can misinterpret directives because it loses the big picture.
Effective Use Cases Today
Rachel: Are there any brands outside of the major AI players that are using this effectively right now?
Vlad: In my field, there is a company, [Vercel], offering a serverless experience where they use AI to find bugs and propose fixes. They do it slowly so it doesn’t destroy your work. Another good use case is for reporting issues or opening help tickets. You might write two or three words, and the AI expands the ticket with more information so the person handling it has less back-and-forth admin work.
My partner, who works in the interior design space, is also seeing use cases in interior design for managing automations and creating images. If you have a house blueprint or 2-D plan, the AI can make a nicer image based on it. These are good use cases, but even building a chatbot isn’t as easy as people think. It depends on the quality you want as an end result.
Vlad Moraru is the technical director at CreativeFuse.