Stop building chatbots for everything

·5 min read

Every AI product pitch deck in 2024 has the same slide: a chat interface with a friendly message like "How can I help you today?" It's the new "Uber for X"—a lazy pattern everyone copies because everyone else is doing it.

Here's the thing: chat is a terrible interface for most problems. But we keep building chatbots because that's what OpenAI showed us, and now we're stuck in a local maximum of unimaginative product design.

Chat is the hardest possible interface

You know what's harder than building a good form? Building a chat interface that has to figure out what form the user is trying to fill out based on their rambling stream of consciousness.

Chat interfaces promise infinite flexibility but deliver infinite confusion. Users don't know:

  • What they can ask
  • How to ask it
  • What format the answer will take
  • Whether their question is even answerable

I watched a user spend 10 minutes trying to get a chatbot to update their billing address. The solution? A form with one field. But no, we had to build a conversational AI that could handle "I moved" and somehow figure out they meant their billing address, not their shipping address, not their office address, not a philosophical statement about personal growth.

The tyranny of the blank text box

A blank chat input is hostile UX. It's learned helplessness as a service. You're making users do the work of figuring out what your product can do, one frustrating conversation at a time.

Compare these experiences:

Chat: "Hi! How can I help?" User: "I need to analyze some data" Chat: "What kind of analysis?" User: "Sales data" Chat: "What would you like to know?" User: [Closes tab]

Dashboard: Three buttons: "Upload Data", "View Reports", "Export Results" User: [Clicks Upload Data] User: [Gets exactly what they expected]

The dashboard is "dumber" but infinitely more useful. Users know immediately what's possible and how to do it.

We're building chat because we're lazy

Chat interfaces feel easier to build. One input box, one output stream, ship it. No need to think about information architecture, user journeys, or interaction design. Let the LLM figure it out!

Except the LLM can't figure it out. It doesn't know your business logic, your constraints, your user's actual goals. So you add more prompts, more context, more guardrails. You implement "conversation flows" and "intent detection" and "entity extraction"—essentially building a terrible form parser with extra steps.

I've seen teams spend months building chat interfaces that could have been replaced with three buttons and a dropdown. But buttons aren't AI enough. Buttons don't get venture funding.

When chat actually works

Chat interfaces work when:

  1. The problem space is truly open-ended. Customer support, therapy bots, creative writing assistants.
  2. Context builds over multiple turns. Debugging sessions, tutoring, iterative refinement.
  3. Natural language is the natural interface. Voice assistants, accessibility tools, command lines for non-programmers.

That's it. That's the list.

Your B2B SaaS dashboard doesn't need chat. Your data analysis tool doesn't need chat. Your PDF processor definitely doesn't need chat. They need thoughtful interfaces that guide users to success.

What to build instead

Structured generation. Instead of "Chat with your data," give users "Generate Report" with smart templates. Let AI fill in the details, but give users control over the structure.

Intelligent forms. Use AI to suggest field values, validate inputs, and provide contextual help. It's boring. It's not sexy. It works.

Augmented workflows. Embed AI into existing interfaces. Enhance the "Export to Excel" button with AI-powered formatting. Add a "Suggest Improvements" option to document editors. Meet users where they are.

Progressive disclosure. Start with simple options and reveal complexity as needed. Use AI to determine when users need advanced features, not to hide all features behind a chat interface.

The real innovation isn't chat

Stripe doesn't have a chatbot for payments. Figma doesn't have a chatbot for design. Linear doesn't have a chatbot for issue tracking. They use AI to make their core interfaces smarter, faster, more intuitive.

The next wave of AI products won't be chatbots. They'll be products that use AI invisibly, products where the AI is so well integrated you don't even notice it. Where the interface guides you naturally to what you want to do, and AI handles the complexity behind the scenes.

Stop defaulting to chat

Before you add that chat widget to your product, ask yourself:

  • Do users actually want to have a conversation, or do they want to complete a task?
  • Is natural language the best way to express this need, or would structured input be clearer?
  • Are you building chat because it's right for users, or because it's expected of AI products?

Nine times out of ten, the answer is to build a better interface, not a chatbot.

The future of AI isn't conversations. It's AI-powered products that are so intuitive they don't need conversations. Build those instead.