Most product managers think adding AI to an app = adding a chat box... but there are so many other ways to subtly enhance your app with AI!
ChatGPT was the first to define the AI UI as a chat box. But we all know deep down that this is not the ideal UI. It's working well for coding agents, but for most normal people, it's tough to stare into blank box and know what to do.
I had this experience with the Muse agent setup the other day. It immediately wanted to connect my Gmail and my Google Calendar and wanted me to come up with a task for it to do. I just closed it because I didn't have a task and didn't want to connect my personal things to it.
As I was imagining Agent Mode for my app, Sanskrit Lexicon (a Sanskrit-English dictionary), I was also imagining a chat interface. But the users of my app are not AI-native (far from it!), so I imagine it would be pretty intimidating for them to use it.
As an example… Last week, I decided to try out Cursor's new project features that I saw being hyped. It looked great! Maybe it was time for me to switch back from VSCode to Cursor for my website code. I switch, and all I see is the black box! I literally could not find where my files were listed. They wanted me to make changes to my website by only prompting the agent without ever even looking at the file structure or deciding anything. Sorry, not for me. I switched back to VSCode immediately.
I know that some people argue that the future is "no interface" at all. But then why are so many people buying the iPhone Duo? And why do I have so much desire to get one despite every rationalization against it ($2000+ for an iPad that folds?!!! I can buy a whole MacBook for that!! I can buy 10 more Claude Pro subscriptions!! So many tokens!!). Meanwhile, I have absolutely no desire for Meta's glasses.
So while I still want to get to some version of "Agent Mode" in my app, what's been actually happening is that I've been building agent workflows in the background instead of inside the app. For example, I recently build a full agent workflow to change the word search in my app to a complete multilingual natural language reverse search that nobody else has done before. The user doesn't care that the app is using an agentic workflow to complete the search. All they know is that they search for "the goddess of wealth" and it correctly shows Lakshmi as the top search result.
In my Sanskrit Keyboard app, I trained a small on-device model to predict the next character as the user types. Again, the user has no idea that there is an AI model inside their app. All they know is that when they type, they can click one of the three possible next character buttons to complete their word instead of typing it.
Today, I got inspired to use open-source Sanskrit TTS models to generate pronunciations for each dictionary word. The user will again have no idea that this is AI-generated or the super cool process I went through testing out different models, selecting just the right voice, handling exceptions, etc. All the user will see is the sound button they can click to hear the pronunciation.
So the interface doesn't have to go away, but instead it can be ENHANCED with AI. Adding AI to your app doesn't mean just slapping the ✨ symbol somewhere that pops up a chat bot. Most of us feel super cringe in websites that do that! Instead, AI could be subtly intervened with the existing UI and interface without screaming about it to the user. You AI-enhanced app should "just work."