AI is still taking the tech world by storm, showing up everywhere from personal assistants to self-driving cars. But as we integrate AI into more products, it’s crucial to remember one thing: good AI product design is really just good product design. The basics of creating user-friendly, valuable products don’t change just because AI is involved. I spend time week over week finding good ones, so let’s dive into what I’ve learned, and how this translates for your SaaS products that really means.

One of the biggest mistakes in AI product development is getting caught up in the technology itself. AI is cool, no doubt, but it's a really really shiny wrench that all the media is fixated on. The value of a wrench is determined by what problems it solves. Instead of jumping straight to, “How can we use AI in our product?” The better questions are, “What do our users need, and how can we help them?” THEN ask: “Is there a creative way we can answer that problem while breaking conventional design expectations?”

By focusing on what users actually want and need, you can make sure that AI is used in a way that adds real value; aaaaaand that’s how you get something more than a cool tech demo!

Just because AI can handle complex tasks doesn’t mean the user interface should be complicated. (Non-developer) Users shouldn’t have to be tech experts to use a product effectively. My rule of thumb looks something like this:

At no point during SaaS onboarding does a client need to know about the intricacies of the export settings. Those settings ARE important; Though All things in due time. Discussing all the individual levers for AI tuning is NOT needed to get most user tasks done. Please give your users simpler options for interacting with the product.

Show users what they need to know WHEN they need to know it. How does all that translate to AI-powered products? The goal is to make the AI’s capabilities easy to access, not to show off how complicated it is behind the scenes. If you can abstract the AI function, I vote you do it. If you can’t for a variety of reasons, contextualize the inputs you need via humanly digestible information. No amount of tooltips filled with tech jargon will fix that issue - what you put IN the tooltip can help. 

AI makes a lot of decisions, from suggesting what movie to watch to automating important tasks. For users to trust these decisions, they need to understand what’s going on. Good design means being transparent—explaining what the AI is doing (in humanly understandable language) , why it’s doing it, and how it affects the user. When users know what’s happening and why, they’re more likely to trust and stick with the product.

Let’s work with an example of a menu planning product: “Now crunching your answers” (What it’s doing) followed by a “Fetching recommendations” (Why It’s Doing It) , “plating your suggestions” (The net impact of these decisions - dressed up as serving results). Your outputs are the suggestions, and when possible, presenting some kind of rationale behind the inclusion. In the context of LLMs it’s possible to squeeze this info out of your prompt responses. In non-LLM contexts, a simple: “This recommendation is 80% a good match”.

Translating a technical, model-assessed confidence score into human context is key to helping later with Machine Learning Ops work, QA work, and opens you a path for public performance improvements. It’s about giving users enough info to feel confident in the AI.

Understanding, and outlining how users currently do perform tasks within your software (and others) is a huge part of usual UX design steps, and AI Features are no different.

AI should make things better, not harder. Good product design means that your new AI features will need to fit smoothly into existing workflows and systems…well... Unless you’re refactoring your SaaS by cutting entire chunks out of your workflows, and slimming your systems thanks to AI. If your SaaS has an API, or plays nice with other systems, it’s definitely in your interest to evaluate & make decisions on whether your shiny new AI functions should join the fun and be on its best behavior. A good AI feature should be part of a bigger ecosystem of features, helping users accomplish their goals without making you refactor your entire client-facing API.

Similarly, you can extend the same intent on the product scale as well. If your AI powered product offer radical alternatives to existing solutions, you’ve got to frame your value proposition around that, and the (im)material benefits (Selling on outcomes over outputs any dayyyy 🤘 ) you’re putting in front of the users.

AI brings a lot of new possibilities, but the basics of good product design don’t change. By focusing on user needs, being transparent in-experience, and fitting your improvements with the right context into existing systems & paradigms, we can create AI-powered products that truly deliver value. In the end, good AI product design is just about good design, period.

Here’s an anonymous channel for you to send me your thoughts if the comments section isn’t for you! 😉