One of the most limiting recent UX patterns is the AI button. You’ve seen it everywhere: a magic wand or a sparkle icon with vague descriptors like AI Companion, Enhance, or Generate.
This UX pattern introduces two key frustrations:
- First, it’s often unclear what the button will actually do. You may have a small text box to add a user prompt, but you’re at the mercy of the quality of an opaque system prompt.
- Second, this approach introduces unnecessary cognitive load by creating an artificial distinction between AI-assisted and manual workflows. When LLMs were still nascent and we didn’t fully trust their output, it made sense for users to switch between their primary task and AI assistance consciously. Today, as models have dramatically improved and we’ve grown more comfortable with AI integration, this forced context switching unnecessarily fragments the user experience.
We need a better alternative to the AI button that is more contextual and doesn’t impose an artificial boundary around AI capabilities. An emerging pattern I really like is the shadow teammate: a live collaborator that exists within your workflow. It turns out to be a good UX pattern for a wide range of tasks, from design to coding. A few good examples:
- Krea Realtime image gen: Krea displays two canvases side by side. On one, you draw whatever you want, as casually as you want. On the other, an image generation model renders your drawing in realtime, guided by a prompt matching your desired aesthetic. AI image-gen in Krea isn’t a separate function, but a parallel, collaborative process.
- Teammates: Teammates spawns AI coworkers that live in Slack. You @mention them just as you would a human colleague. There’s no “AI mode”. You simply communicate with these agents through the same channels and conventions you use with human teammates. You can give them tasks, ask for updates on ongoing tasks, or, my favorite, banter with them.
- Github comments: Several AI coding tools comment directly on pull requests, using the same format and conventions as human reviewers. It’s a natural extension of existing code review conventions, complete with all the back-and-forth needed for clarity.
What makes these examples work? In each case, there’s a natural back-and-forth like you would have with any coworker. You’re able to provide enough context to overcome a bad system prompt. You get quick feedback and can iterate quickly on a goal together. There’s no mysterious Enhance command or separate AI mode to switch into. It just … works.
The AI button feels like a transitional design pattern. It’s this generation’s Clippy, and it’s time for it to go.