This week's edition covers building custom interfaces in ChatGPT, Google's Veo 3.1 video generation with native audio, multi-turn agent evaluation, and monitoring agent reasoning.
Apps SDK Brings Custom UI to ChatGPT | Build AI for December 2, 2025
0:00
/657.30424
Transcript
OpenAI: apps inside ChatGPT
OpenAI has just launched something called the Apps SDK, and it's a bit like giving developers a new set of building blocks for ChatGPT. Instead of just chatting, you can now create apps that live right inside the conversation, with their own custom look and feel.
🐦⬛
We've collectively decided that chat windows are the new operating system—because why would we want dedicated interfaces when we can cram everything into a text box?
The SDK builds on an open standard called the Model Context Protocol, which is basically a way for ChatGPT to talk to other tools and pull in outside data. Now, it's not just about connecting things up – you can actually design how your app works and looks, all within the chat itself.
The Apps SDK is open source, which means anyone can look under the hood and even build their own versions. In theory, these apps could run anywhere that supports the same standard, but that depends on whether other platforms decide to join in.
🐦⬛
Every platform will promise compatibility while secretly hoping everyone commits exclusively to their version first—the prisoner's dilemma, but make it tech standards
One of the big draws is that you don't have to hand over your whole system to OpenAI – you can keep your own logins, your own premium features, and connect straight to your own servers.
🐦⬛
The fact that "you get to keep your own data" counts as a selling point tells you everything about where we've ended up
Plus, your app can remember what's happened in previous chats, so conversations can actually build up over time.
If you want to get started, OpenAI has put out some design tips and a public codebase with examples to play with. Right now, you get the basics, but more building blocks and quicker ways to put things together are on the way. There are still some big questions, though: we don't know exactly what UI tools or performance tricks will be available, or which kinds of apps will really shine inside a chat window. It's early days, and a lot is still up in the air.
So why does this matter? If you're building anything with AI, this is a whole new way to get your app in front of people. It's a bit of a crossroads: do you stick with the old way of making apps, or do you try blending chat and interactive features together? The open standard is meant to keep things flexible, so you're not stuck with just one company's system – but that only works if others decide to play along.
🐦⬛
And by "play along" we mean "adopt our standard instead of creating seventeen competing ones"—because that's never happened before in tech history
One of the big upsides is that you keep the keys to your own house – your servers, your user logins, your premium features. OpenAI says they'll add ways to make money from your apps soon, including sharing revenue and letting people buy things right inside the chat.
🐦⬛
Monetization features coming "soon" is the tech industry's version of "the check is in the mail"
But remember, this is still a preview. The tools and guides will change, and if you jump in early, you might get a head start before everyone else joins the party.
🐦⬛
Nothing says "stable platform decision" like building on tools that will definitely change before launch
Key evaluation questions for developers:
Does your application benefit from conversational invocation?
Can you design effective workflows that blend chat and interactive elements?
Is it worth putting in the effort now, knowing the platform is still new and not even available everywhere yet (like in the EU, Switzerland, or the UK)?
🐦⬛
That last one revealing we've reached the point where regulatory fragmentation means building for "global" actually means "some of the globe, details TBD"
More AI everywhere doesn’t mean better results. Small teams of AI experts beat mass licenses. Attackers now use AI to run cybercrime at machine speed. Leaders can’t keep up. Tools still forget us. The winners? Focused expertise and hybrid AI, not AI everywhere.
You've built something and you need to know if it works. So you do what's sensible—you ask an LLM to grade it. Factual accuracy, code quality, agent outputs. The machine judges the machine, and you get a number you can act on. Except that number
Imagine you have a bunch of teams, some with AI, some without, and some where everyone gets their own AI. Researchers ran a big experiment with over 400 people to see what actually happens when you mix and match humans and AI in different ways.
Here’s what they found:
