Amelia Wattenberger on AI UX Beyond Chatbots

Table of content

Amelia Wattenberger designs AI interfaces that respect human cognition. As a Principal Research Engineer at GitHub Next (now building at Sutter Hill Ventures), she’s spent years prototyping novel UIs that combine AI capabilities with thoughtful interaction design.

Her core argument: chatbots are a lazy default, not the future of AI interfaces.

The Problem with Chat

When you open ChatGPT, you see a blank text box. No hints about what works. No guardrails. No affordances.

“Good tools make it clear how they should be used,” Wattenberger writes. “A good pair of gloves is hand-shaped. Metal mesh prevents physical harm, rubber prevents chemical harm. A chat interface looks the same as a Google search box, a login form, and a credit card field.”

Every prompt is a pile of context that users must manually assemble. Who are you talking to? What format do you want? What should be avoided? These questions get typed again and again, when they could be baked into the interface.

Her fix: add controls. On Copilot for Docs, her team added sliders for experience level and detail preferences. The interface captures context that users would otherwise hack into each question.

The Isolation Problem

Chat responses exist in isolation. Ask an LLM to improve your writing, then ask for more active language, and you’re stuck scrolling between responses, comparing line by line.

“As someone always thinking about how AI can help edit code or prose, I can’t help but see the inability to have a ‘working buffer’ as a complete non-starter.”

This breaks the natural creative loop. A painter steps close to smoosh paint, then steps back to evaluate. Good tools let you choose when to switch modes. Chatbots force constant switching—ask, wait, read, repeat. No flow state possible.

LLMs as Thinking Partners

Despite her chatbot critique, Wattenberger sees massive potential in LLMs as tools for thought—not for generating answers, but for offloading cognitive work:

Memory extension: “When talking through an idea with an LLM chatbot, I can trust that it will remember all of the context. I’m happy to throw an idea out there and quickly move on to the next one, knowing that it will be waiting for me.”

She tells the AI to “pin this idea for later” and retrieves pinned concepts weeks afterward.

Suggestion generation: A thesis needs supporting points. LLMs can suggest related ideas, propose analogies from different domains, even evaluate whether an analogy strengthens or weakens your argument.

Perspective shifting: Like having someone stand back from your canvas while you paint. She asks about the “flow” of writing as she tweaks it—getting a zoomed-out view that’s hard to achieve when reading word by word.

The No Man’s Land

Wattenberger warns against the middle ground where humans remain responsible but aren’t in control:

“When a task requires mostly human input, the human is in control. But once we offload the majority of the work to a machine, the human is no longer in control. There’s a no man’s land where the human is still required to make decisions, but they’re not in control of the outcome.”

The goal: tools that amplify human abilities, not machines that turn us into operators pressing buttons.

Projects That Show the Way

Her work at GitHub Next demonstrates these principles:

Code Brushes: Paint code with AI, selecting regions and applying transformations like an image editor
GitHub Blocks: Composable, customizable views of repository content
Copilot for Docs: Chat with documentation, but with controls for experience level and context
Code Atlas: Combine LLM reasoning with rigid structure for more robust responses

She also builds personal AI tools: brainstorming interfaces for organizing thoughts, writing assistants that suggest improvements rather than replacing your voice.

The Invitation

“I invite you to try it. Instead of using chatbots to steer toward specific answers, try using them as partners in that natural cycle of taking in, transforming, and sending out that drives creative thought.”

The chatbot is a 20-page restaurant menu—good enough for anyone, ideal for no one. The next generation of AI interfaces will be purpose-built tools that fit specific workflows and respect human cognition.

Wattenberger is building them.