Don't just slap on a chatbot: building AI that works before you ask

You know what's wrong with AI today? Everyone's racing to add chatbots to their products, just slapping on a chat interface and calling it a day. And I get it, it's the easy solution, but is it actually helping anyone?

Hi, I'm Arthur, I'm a product designer at Evil Martians, focused on dev tools and AI. Today I want to challenge how we think about AI in products, share what I've learned from the trenches, and give you practical principles you can apply in your own work.

When I started working on Tegon, an AI issue tracker, I decided to do something different. Actually, I did the complete opposite of what everyone else was doing. And you know what? It worked. Here's something funny: the solution we needed has beenright in front of us for 25 years.

Remember Clippy? Yeah, everyone loves to hate on Clippy. But hear me out: Clippy had theright idea, just terrible execution and terrible timing. But now we have the technology to actually do itright. While everyone else is making their AI assistant more reactive, I thought, what if AI didn't wait to be asked?

What if it just knew what you needed? Let me show you what I mean. Watch this. See what's happening here? Our AI didn't wait for user input; it saw someone reporting a legal issue and immediately started asking theright questions.

Non-generic "how can I help?" questions. Specific, contextual questions that actually move the work forward. We call this "suggestion mode." It's tracking what the user is writing in real time, understanding the context, and jumping in at exactly theright moment, with exactly theright questions.

No chat window needed. This next example is even better. Look at what happens when someone writes an issue that can be split into sub-issues. Most AI assistants would just sit there, hours since the complexity, and immediately suggest a better way to organize the work.

See how it's already identified the sub-issues? And it's not just splitting things randomly. AI has all the previous data, knows how it was done before, and has a general understanding of how it is better to organize things. This is AI that actually understands your domain.

We call this "action mode." And here's where it gets really interesting. Our AI isn't just organizing work; it's thinking about timelines and resources. It's like having a really good project manager who's always there, always paying attention (if those even exist).

Notice how all of this happens within the natural flow of work. No context switching, no extra windows, and no chat interfaces. It asks you theright question and helps you to better manage issues. We call this "question + action mode."

Now, I know what some of you are thinking: "But what about user control?" Well, you can revert the changes easily with one click. You're still in charge. So what you have just seen is AI that guides users seamlessly.

No more wasting time trying to figure out how to explain what you need. The AI has your back, helping you create better work without breaking your flow. And while we build this for issue tracking, think about how powerful this pattern could be across all kinds of professional tools.

We follow three simple rules to foster this sort of proactive AI in our products. First, AI should supplement user agency, not replace it. Second, AI should offer recommendations, never force them. And third, AI should be part of the natural workflow, not stop it.

Take code editors, for example. AI can proactively watch for common pitfalls and suggest improvements. This would be especially valuable for developers just learning a new language or a framework. Or design tools: imagine a tool that makes suggestions toward accessible design as you work, so you don't have to check it afterward.

That's proactive AI in action. And communication tools could prepare relevant context before your meeting even starts, or find you the document you just mentioned on a call. You're still in control, but now you have an advisor in your corner.

Here's how you can start thinking about this for your own products. First, look for friction points where users have to stop what they're doing to ask for help. That's your opportunity for proactive assistance. Second, identify patterns in your users' behavior.

Where do they usually need help? What questions do they always ask? These are your clues for automation. And finally, the most crucial: think about context. Where do users get stuck? That's where AI can help most. Look, we can keep adding chat interfaces to everything.

AI interface design is still just beginning to take shape. That means we don't have a fully formed playbook of best practices just yet, but simply copying and pasting chat interfaces isn't the answer. So, don't be afraid to experiment in your products.

Challenge the status quo, and propose unexpected UI solutions. I'm excited to see what you will build with these ideas. Thank you for your time. If you want to know more, make sure to check out our blog and social media.

Bye!