Key Takeaways

  • ChatGPT Sites includes integrated Cloudflare D1 SQL databases, Cloudflare R2 object storage buckets, and multi-user co-editing out of the box.
  • The platform hosts Model Context Protocol (MCP) plugins directly, allowing developers to run backend connectors without maintaining separate server infrastructure.
  • User behavior is shifting from building persistent web properties to deploying disposable software, like single-week custom interfaces for business trips.
  • Teams are adopting Sites for non-publishing workloads, including hosting custom data, serving internal plugins, and running model evaluations.

The Disposable Software Shift

For decades, writing software carried a steep carrying cost. If you wrote code, you had to provision servers, configure domains, set up databases, and commit to maintenance. Because deployment was expensive, software had to be permanent. You only built an application if you intended to maintain it for months or years.

OpenAI Product Lead Kath Korevec points out that coupling Codex with ChatGPT Sites breaks that economic rule. When deployment overhead drops to zero, software becomes disposable.

“I might just want like a different view of Google calendar for the day or for the week,” Korevec explained. “I might be like going on a on a business trip and I just want like things organized differently. And so I'll build a site for that week and then I'll throw it away at the end of the week.”

This changes how technical founders should think about tooling. Instead of cramming workflows into rigid enterprise software or spending days customizing third-party dashboards, you can generate a bespoke interface for a three-day project, use it once, and delete it. The application does not need documentation, a migration plan, or an upgrade path. It only needs to solve the problem directly in front of you right now.

Full-Stack Plumbing Behind Sites

Most developers initially assumed ChatGPT Sites was simply a static host for HTML prototypes. In reality, the runtime includes real backend infrastructure.

“One of the reasons why we built sites is because we wanted a place to basically put software someplace,” Korevec said. “It does sites does come with D1 data. It has an R2 bucket. It's all deployable. You can have co-editors on it.”

By packaging Cloudflare D1 (serverless SQL) and Cloudflare R2 (object storage) directly into the environment, Sites handles stateful web applications without requiring developers to wire up external cloud accounts or configure API credentials.

OpenAI expanded this backend capability by introducing hosted Model Context Protocol plugins. As Korevec noted: “We announced today that you can actually host MCP plugins through sites and then use those for your sites.”

Because of this built-in capability, teams are using Sites for workflows far beyond client-facing web pages. “What we saw was that people were using sites for yes building websites but then also building plugins and hosting data and running evals,” Korevec said.

When your code generator owns the hosting layer, the database, and the protocol plugins, the line between prototyping and production infrastructure disappears.

What to Do With This

Pick one repetitive, messy workflow you are managing this week, such as filtering inbound sales leads from a CSV or comparing outputs from model prompt tests. Open Codex, instruct it to create a custom single-page tool connected to a ChatGPT Site with local state, and use that micro-app exclusively for the task. When the task finishes on Friday, archive the site instead of maintaining it.