By my reckoning, this AI experiment is 176 years old. Scratch that—three years old. Some memories still leave me wondering whether they’re mine or stories my dad told me as a kid.
Even after three years, chat remains the primary interface for LLMs. I’m pretty sure @pmarca argued for a textarea component in the HTML spec. I know because I asked AI about it. In a textarea.
Steven Pinker expressed it better than I can: it’s a shame the first large-scale AI implementation was a gimmick, a first-person chatbot. The real promise is task-oriented AI. And when an academic says it, and I quote it in a post, it must be true.
Chat became the default because it was the first thing people clicked with. It offers a universal solution precisely because nobody knows what users will throw at it. But you know how you will use it—and that is exactly when chat fails you.
What you actually need is a customizable UI you can manifest on demand. In GitHub Copilot, that is called a canvas—a full-stack app that runs inside the Copilot app with zero browser chrome. The agent talks to the server side, the server responds, and the resulting surface can do anything a normal program can while communicating bidirectionally with the agent.
That sounds hand-wavy, so here is a concrete example.
I built a Connect 4 game as a canvas where I could play against the agent inside Copilot. On high reasoning, I absolutely destroyed GPT-5.6 Sol. Then GPT-5.6 Luna beat me with no reasoning at all. Listen, Connect 4 is a hard game.
Creating one is as simple as asking:
Create a new canvas that uses the Connect 4 game to demonstrate the ability for the user to interact with the canvas for the canvas to talk to the agent and for the agent to control the canvas
The app already understands what a canvas is. No further explanation is needed.
Because canvases are real full-stack apps, not merely web pages, they can call third-party APIs and execute code locally on your machine. I built a Winget UI that browses the registry and manages local packages—install, uninstall, the lot. No AI was involved, and that is the entire point.
Chat pressures you to use the agent for everything, which is just burning tokens. It is almost always better to have the agent build a tool where future interactions are free than to treat the agent itself as the tool. Stop asking GPT-5.6 Sol Max to "stage and commit" every time. I know you do it—I do it too. Please spare me the token-shaming.
The same principle applies to SQL workflows. Instead of asking chat to run every query, have it build a proper query UI once. Every query after that is instant and free.
Next Coding agents actually make software engineering management more difficult →
All Replies （1） #
Want a live back-and-forth? Join the global AI chat room — login to talk.
The author’s 176-year vs. three-year slip nails how AI warps time perception—memory blurring isn’t just poetic, it’s cognitive drift from constant synthetic interaction.