Here's how Gemini 2.5 Flash-Lite writes the code for a UI and its contents based solely on the context of what appears in the previous screen - all in the time it takes to click a button. 💻 ↓

So it just randomly makes shit up? That seems kinda useless. I have apps and files because they do the same thing every time I open them.

Imagine wasting token $ to navigate your filesystem 🥲.

Imagine replacing your wife's OS with this.

built a prototype similar to gemini-computer with Gemini 2.5 Flash-Lite, checkout and enjoy 🎉

Another flash lite demo

This is coming faster than I thought !

The idea of a dynamically-generated UI that's appropriate specifically to that context is a great one. But this demo is horrible and it doesn't do anything to illustrate the promise of such a UI. The only thing that you've shown here is inconsistency in simple basic interfaces.

It’s @websim_ai but less cool

Here's a different take on using Gemini 2.5 Flash-Lite for ad-hoc UI generation:

automatic GUI generation requires accessibility and consistency considerations, the SUPPLE project linked above had been considering this problem for years

Wild how we’ve gone from “autocomplete” to “auto-interface” 😳

The UI will become very fluid and very personal very quickly. Crazy times ahead.

I think this is great. The current concept of a folder is to store static assets, but if you think about why we’re storing static assets, it’s to share context or help us remember it. Think you are an hardware engineer designing a PCB or mechanical part, you go back and forth between design software, folders, and productivity apps. A directory that understands your current context and gives you only the information you need like an IC pinout or issues from Slack, Jira, or Confluence related to the hardware you’re designing would be perfect. That’s your ultimate intent for storing files and using these tools. This directory could capture all of that, which is amazing. We are trying to implement similar on Partix!

A world where you can instantly customize your UI means designers can collect more data on how users want the app to look at feel.

Fever Dream OS

Fucking shit they did it they actually got ahead of it MAD LADS

Excited for what's to come. Especially with the gemini diffusion model, once ready.

massive. text diffusion will be big, can't wait to have access to the API!

oh my god you just showed me how we can plug into the matrix

Have you all looked at your diffusion model for doing this? It's so much faster than anything out there that it seems ideal for this type of work. Then again it's definitely not all that smart so maybe not quite yet :)

The value of Flash-Lite is in heavily embedded usage, empowered by lots of prompt stuffing with relevant content to overcome the lack of world model capabilities, strong guardrails and doing that at scale and reliably. Not making random shit.

This is 🤯 So front end devs and designers are cooked now

To give realistic @Google feel add the first 4 icons as sponsored links.

@websim_ai was first and better :) @RobertHaisfield

Cool! IMO generative interfaces will progressively deprecate rigid interfaces. That's been Widgens' vision since day 1. Check out this (even cooler) demo of a generative interface

Gemini is cool and all, but that UI? Straight outta the golden era of Google….. I need it!

All we need is for the Start menu to be rendered on demand in a LLm running on 100GPUs in the cloud. 😂

Gemini 2.5 🔦

UI code in a click is wild. Frontend starting points are looking better and better

Ah yes, who doesn't like unexpected changes every time they use something?

This is what future-first development looks like. #Gemini #UIUX

Gemini 2.5 Flash-Lite is impressive tech. It's fast, it's context-aware, and it'll definitely replace that one guy on your team who only knows how to center divs. But until it can explain why the CSS is broken in Safari, I'm keeping my job.

Code?

Incredible. What is the impact of not needing to write, store, load and maintain programs? What time and resources are freed up? What can be done with those recoverings? What does it mean to have an LLM at the core of an OS?

looking good. wonder if it has access to a set of prebuilt components it's putting together each generation. also does it remember what you did in previous screens?