Andrew Ambrosino on Fast Models and Dynamic UI Generation

Key Takeaways

• OpenAI rewrote ChatGPT.com in five weeks to position it as an executive app capable of handling agent workflows across desktop and cloud.
• Products like Dots, Codex Cloud, and ChatGPT Work run centrally in the cloud while retaining direct access to your local machine when required.
• ChatGPT Space replaces static back-and-forth chat threads with multiplayer workspaces that support live artifacts and data visualizations.
• Ultra-fast model inference creates a qualitative leap in product design by generating bespoke, ephemeral user interfaces on demand.

Cloud Agents That Touch Local Machines

Most software forces a choice between desktop utilities with file access and cloud apps built for collaboration. The engineering team at OpenAI decided that boundary makes no sense for autonomous agents.

Andrew Ambrosino explained how the team unified these environments across products like Dots, Codex Cloud, and ChatGPT Work. As Ambrosino put it: “One of the things that we really like about all of our new products today, whether that's the dots, the Codex cloud, chat work, all of this stuff is running in the cloud and has access to your local machine if it needed it.”▶ 48:42

Instead of bouncing between scattered windows, agents run centrally while executing commands directly against local file trees. Ambrosino noted the internal pain that drove this architecture: “We find ourselves bouncing around between applications all day, right? We're you know in this in the Codex desktop app, chat desktop app. All sorts of documents were sharing things across Slack and so we wanted to make something that felt very agent native and felt like a way to collaborate with other teammates.”▶ 52:36

To anchor that change, OpenAI rebuilt its core consumer surface at breakneck speed. “One of them is that we completely re rewrote chatgbt.com in the last five weeks. To be the CEO app,”▶ 47:46 Ambrosino said. The rewrite turns chat into ChatGPT Space, a shared canvas where multiple team members watch an agent build live data views, run scripts, and generate interactive components in real time.

Speed Changes Product Architecture

When models run slowly, interfaces default to text boxes and streaming paragraphs. When inference becomes nearly instantaneous, the product architecture shifts toward generating disposable software.

Ambrosino observed that raw latency reduction mimics a sudden jump in cognitive capability: “Ultra fast has been one of those things that for us like you just you can't go back, right? There's it's, you know, it's not exactly the same as intelligence doing something that quickly, but it sure feels like it when you, you know, especially when we do these dynamic UI things.”▶ 56:12

Instead of asking users to click through pre-built forms and rigid dashboards, the interface creates bespoke micro-apps for the exact problem at hand. If a user needs to compare five financial scenarios, the system renders a custom interactive calculator rather than dumping numbers into raw text.

The real challenge for founders is keeping pace with the underlying model capabilities. As Ambrosino admitted: “Sometimes I think that the the models are changing so fast that it's hard to get the product timing on these things exactly right to the model when the model is best at it.”▶ 52:05

What to Do With This

Audit your product's core workflows by Friday and identify where you still return static text responses to user prompts. Replace at least one conversational bottleneck with an interactive, generated artifact or component that renders immediately inside the user's workspace.