Amelia Wattenberger said the text box was a dead end while the whole industry was shipping text boxes, and built the argument so you could operate it yourself

In 2023 every product on the internet grew a chat box. It was the shape the entire industry agreed AI came in: a rectangle, a blinking cursor, and the instruction to describe what you want in words.

Amelia Wattenberger published an essay arguing this was a dead end, and the case she made was not aesthetic. A text box tells you nothing about what the system can do, gives you no way to adjust anything by degree, and forces you to guess the magic words. We spent forty years building sliders, toggles, direct manipulation and spatial arrangement precisely because those things carry information a sentence cannot, and then we threw all of it away for a prompt.

What makes the piece land is that it is not written, exactly. It is operable. The controls are in the essay. You push a slider and feel the difference between adjusting a thing and describing an adjustment, and the argument stops being a claim you can disagree with and becomes an experience you just had.

The hard part is that she was early

Anyone can write that a popular thing is bad. The difficult version is writing it while the thing is ascendant, in front of an industry with enormous incentive to disagree, and being specific enough to be provably wrong later. She named the mechanism, not the vibe, which is what let the essay survive three years of the field moving.

It is a design engineer's argument in the strictest sense: only someone who both designs interfaces and builds them could have made it, because the case rests entirely on knowing what a well-built control actually does for a person. That combination is why the same essay gets cited by designers and engineers who otherwise cite nothing in common.

Now she is building it

She is designing Intent at Augment Code, a workspace for running AI coding agents, and it is recognisably the essay in product form. The system ships with specialist roles, a coordinator that delegates to implementers and verifiers, and all of it adjustable rather than incanted. The through-line she keeps naming is that the specification becomes the source of truth and the work moves from implementation to intention.

There is a longer history behind it. She wrote Fullstack Data Visualization with D3, which taught a generation of engineers to build charts properly, and did research and development at GitHub Next on exactly the question of what developer tools become. She also supports companies at Sutter Hill Ventures.

Three years ago, saying the chat box was a dead end was contrarian. Now the interesting products are the ones that agree with her, and she is building one of them. Read the essay first, and use the sliders.

Why chatbots are not the future.

A 2023 title that read as contrarian and now reads as early