AI is having the wave of the decade in both consumer and business software. If you're even somewhat online, it feels like a new AI powered app drops every day, with no end in sight. And while new tools are extremely exciting, it got me thinking about the bigger picture of AI powered apps: are we really implementing AI the right way?
In all honesty, this question first came to mind while we were working on our first email client: Zero. We wanted to implement AI (like every founder does) with the end goal of saving our users time in their inbox. However, it felt like every AI feature we added, the harder and less enjoyable it was to use the app.
You might then be asking, "how does adding AI into an app make it harder to use?". This came down to one issue: the way users interacted with the AI was suboptimal and quite unenjoyable. It added friction instead of removing it.
So what were we doing wrong? We were thinking of AI implementation like a developer instead of an industry expert.
The problem: "Cursor for [insert industry]"
Right now, tons of new and existing apps have the "AI-powered" label. However, that typically translates to a simple chat box that spits out text, usually in the form sidebar / chat interface. In my opinion, this form of AI interaction can help in some contexts, but is it really the best pattern for user workflows? I'd argue not.
So, if AI isn't optimal in this chat interface, why does everyone do it? The answer is actually quite simple: People building AI applications (developers) saw the success of Cursor (an AI powered code editor that uses a sidebar to communicate with the agent) and figured that it's the most optimal way of using AI in every other app.
That is exactly what we did with the initial versions of Zero. We simply added a chat interface into our app, an AI composer section, a few tool calls to add labels, and called it a day. There was no true time savings or productivity in what we built. Our users got no benefit.
This was a 2D way of thinking: copy a pattern that works (Cursor and ChatGPT) and hope it works everywhere else. I also saw this first hand in the X25 YC batch. Lots of very talented founders and developers who had amazing ideas were building with the exact same AI pattern as Cursor did and hoping it would work. The end result? Lots of "Cursor for [insert industry]" instead of tools that actually meet people where they are.
The real goal: fewer steps to the outcome
If AI is going to play a role in everyday software, it should remove steps, not add them. A chat box is fine, but it should be a secondary option. It should be a piece of the pie, not the entire pie. The way to interact with AI should almost feel effortless and invisible, almost like it's reading the users mind.
Here's what this looks like in Orchid:
- AI drafts email in your voice when a new email comes in, all you have to do is approve
- AI automatically shows you what you do and don't care about via previous actions
- Feature that lets you organize your inbox with a click of a button via AI
These actions don't make you "think in prompts." They surface likely intents and put them one tap away. Less "operate the tool." More "the tool already knows what you probably want." This approach cuts friction, lowers brain load, and truly saves time without forcing you to learn a new interface.
Our current stance
Sidebar chat can be helpful, but it often turns into a detour. You switch context, type an instruction, wait for an answer, then copy whatever it says back into your actual task or wait for the implementation by the agent. It might be cool to you but it's not that helpful to your end user.
We're not spiking the football and the solution is still evolving, we are first-time founders with a lot to learn. But our direction is clear: make AI that closes the distance between intent and outcome. Make actions obvious, contextual, and fast. Keep the chat input, but treat it like a power tool, not the default.
Do this well and AI gets quieter and more useful. It feels like the product understands what you're trying to do and just helps you get there faster. this requires more creativity but it's well worth it to the end user.
If you want to jam on what we're building at Orchid and why we think this model actually serves people better, ping me. Happy to share more: nizzy@orchid.ai
