How OpenAI uses ChatGPT Sites (live at DevDay!) | Kath Korevec (Product Lead)

I spent years building websites and apps the hard way. You hire a team, you map out the UI, and you spend months trying to figure out how to pull data from Slack and Notion into a single dashboard that people actually want to use. Usually, by the time the tool is finished, the team has moved on to a different workflow or the API has changed. It is the classic builder's trap: solving a problem with a rigid solution for a fluid team.

At the recent OpenAI DevDay, Product Lead Kath Korevec demonstrated something that shifts this dynamic significantly. They are calling it ChatGPT Sites. On the surface, it looks like just another way to visualize data, but for those of us actually building products, it represents a transition from building 'pages' to building 'interfaces on demand.'

The Death of the Static Dashboard

Most internal tools are where data goes to die. You have your project management in Notion, your fires happening in Slack, and your schedule buried in a Google Calendar. We have spent the last decade trying to build 'central hubs' that attempt to stitch these together. The problem is that a hub built for a developer is useless to a marketing lead. A dashboard built for a founder is noise for a junior engineer.

What Korevec showed is a system that takes these disparate data streams and uses an LLM to generate a functional, live interface. It isn't just a chatbot summarizing your day; it is a site that adapts to the specific user. If you are a founder asking about the roadmap, the site pulls the relevant Notion docs and Slack threads to build a view that matters to you. If your lead dev asks the same site about deployment blockers, the interface reshuffles to show them GitHub PRs and technical logs.

For builders, the implication is clear: we are moving away from the era of fixed UI. Instead of coding every possible state of an application, we are providing the AI with the context and the data permissions, and letting the model decide how to present that information.

How It Actually Functions

The demo focused on the reality of how we work now. We don't work in one tab. Korevec showed how ChatGPT Sites can ingest the chaos of a modern tech stack. By connecting these apps directly, the AI isn't just searching for keywords; it understands the relationships between people, tasks, and time.

From a founder's perspective, this is a massive win for operational efficiency. The time wasted on 'internal alignment'—which is usually just people asking each other where a document is—could be cut to near zero. You aren't building a wiki; you are building a living memory of your company that anyone can query through a visual interface.

The Technical Skepticism

I have to be honest: there is a reason to be cautious here. Whenever we talk about giving an LLM access to Slack and Notion, every security officer in the world gets a headache. OpenAI is positioning this as a secure enterprise solution, but the 'black box' nature of how these models prioritize information remains a concern. If the AI decides a joke in a Slack 'random' channel is more relevant to a project than a formal spec in Notion, the tool fails.

Builders need to look at this not as a 'set it and forget it' tool, but as a new layer of the stack that requires its own kind of maintenance. We used to debug code; now we have to debug context. If the Site isn't giving your team what they need, the problem isn't the CSS or the JavaScript—it's the data quality in your source apps.

What This Means for Founders

If you are starting a company today, your first instinct shouldn't be to build a custom internal portal. It should be to organize your data so that tools like ChatGPT Sites can actually read it. This means moving away from messy, unorganized folders and toward structured, readable documentation.

We are entering a phase where the 'moat' for a business isn't the software they use, but the quality of the organizational data they feed into their AI tools. If your Slack is a mess and your Notion is five years out of date, no amount of OpenAI magic is going to save your workflow. The tool is only as good as the history you give it.

I see this as a massive opportunity for small teams to act like large ones. You don't need a dedicated operations department to build internal tools anymore. You just need a clear head and a well-connected data set. You become a prompt engineer for your own company's productivity.

The Long Game

This isn't just about making Slack easier to read. This is the first step toward the 'Self-Building App.' Today, it is internal sites for teams. Tomorrow, it is client-facing portals that build themselves based on a customer's specific needs. The barrier to entry for creating complex, data-driven software is dropping to floor level.

For those of us in the builder community, this is a call to focus more on the 'what' and the 'why' rather than the 'how.' The 'how' is increasingly being handled by the model. If you can define the logic and provide the data, the interface will take care of itself.

The future of building isn't writing more lines of code; it is providing better context to the machines that write it for us.

I’m staying skeptical on the privacy side, and I’m definitely watching how they handle hallucination in data visualization. But as a founder who has wasted hundreds of hours on internal dashboards that no one used, the promise of a site that builds itself around my team's actual needs is too big to ignore. It’s time to start cleaning up your documentation—the AI is ready to read it.

Read the original at Lenny's Newsletter →