This was a week that matters โ€” not just for Codex users, but for anyone watching where the AI industry is headed. OpenAI filed confidential paperwork for what could be the largest tech IPO in history, shipped a feature that turns natural language into deployed web applications, and quietly drew a line under three platform tools that didn’t make the cut. Here’s what you need to know.

OpenAI Files Confidential S-1: The IPO Clock Starts Ticking

On June 8, OpenAI announced it had confidentially submitted a draft S-1 registration statement to the SEC. The company’s public statement was characteristically candid: “We expect it to leak so we’re just announcing it.”

The filing, reportedly led by Goldman Sachs and Morgan Stanley, positions OpenAI at an $852 billion post-money valuation โ€” with some reports pegging the target as high as $1 trillion. Timing is intentionally ambiguous. OpenAI stated it “has not decided on timing yet” and that there are initiatives better pursued as a private company, but the filing preserves the option to go public sooner if conditions align.

Why this matters for engineering leaders: The S-1’s public version โ€” likely August or September โ€” will be the first time OpenAI opens its books to regulatory-grade scrutiny. Revenue run rate ($14B annualized as of February), gross margins, compute spend, and customer concentration will all surface. For teams building on OpenAI’s platform, this transparency will be a double-edged sword: better financial stability signals, but also sharper scrutiny on pricing leverage and dependence risk.

Codex specifically crossed 5 million weekly active users in this window โ€” a 6x increase since the desktop app launched in February. Non-developers now account for ~20% of users and are growing more than 3x faster than developers. That stat alone tells you where OpenAI is positioning the platform.

Codex Sites: The Prompt-to-App Pipeline Is Real

The week’s most impactful product move was Codex Sites, launched June 2 in preview. It’s exactly what it sounds like: describe an application in natural language, and Codex builds, deploys, and hosts it on OpenAI-managed infrastructure โ€” returning a production URL with authentication baked in.

Under the hood, this is not a static site generator. Sites deployments compile to Cloudflare Worker-compatible ES modules with a managed D1 relational database and R2 object storage wired in. The workflow is deliberately two-stage: save creates a deployable artifact linked to a Git commit for review, and deploy publishes to production. Every Site URL is a production deployment by default โ€” there is no staging tier.

Current access limitations: Sites is in preview for ChatGPT Business (enabled by default) and Enterprise (admin-enabled via RBAC). Plus, Pro, and API-key users are excluded for now, though OpenAI has signaled broader availability later.

Where this fits in your stack: For internal tools โ€” dashboards, project trackers, knowledge bases, operations boards โ€” Sites removes the entire CI/CD pipeline from the equation. Teams that currently burn engineering cycles on CRUD apps for internal stakeholders should evaluate this the moment their workspace has access. Just don’t treat a generated app as production-grade software without a real review gate. The save-before-deploy workflow exists for a reason.

Six Role-Specific Plugins: Codex Expands Beyond Engineering

Alongside Sites, OpenAI shipped six role-specific plugins: Data Analytics, Creative Production, Sales, Product Design, Public Equity Investing, and Investment Banking. Five more are on the roadmap, including Marketing Strategy and Legal.

These bundle 62 popular app integrations and 110 automated skills. The plugins are live in Codex for supported regions and signal a deliberate pivot: Codex is no longer pitching itself as just a developer tool. With non-developer users growing three times faster than developers, the platform is betting that knowledge workers will drive the next wave of adoption โ€” and the plugin architecture is how they’ll serve that audience without fragmenting the product.

The Deprecations That Deserve Your Attention

On June 3, OpenAI announced the end of three platform features, all with a common shutdown date of November 30, 2026:

  • Reusable Prompts (v1/prompts API): Server-side prompt storage is being decommissioned. Migrate prompt content into application code by November 30.
  • Evals Platform: Dashboard and API-based evaluation runner goes read-only October 31, shuts down November 30. OpenAI is directing teams to migrate to Promptfoo as the supported replacement.
  • Agent Builder: The low-code UI for composing Responses API agents is being retired. Migration paths are the OpenAI Agents SDK (for developer teams) or ChatGPT Workspace Agents (for no-code enterprise use).

Read this carefully: The underlying Responses API is not deprecated. Only the builder UI and saved-agent objects are being retired. If you built agents in Agent Builder, export your configurations and port them to SDK code. This isn’t an emergency, but with a five-month window, it’s easy to defer โ€” don’t let it slip to October.

Infrastructure Updates Worth Knowing

Amazon Bedrock support (June 1): Codex is now available as a model provider on Amazon Bedrock, including GovCloud regions. Teams on AWS can route Codex through their existing authentication, account controls, and billing. For enterprise accounts with AWS commitments, this simplifies procurement significantly.

Computer Use on Windows (May 29): Codex can now see, click, and type in Windows desktop apps, matching the macOS Computer Use capability that shipped in April. Remote control support means you can initiate work from iOS and monitor it on Windows.

Codex reached 0.139.0 this week, adding standalone web search in code mode, richer MCP tool schema support, smarter plugin marketplace listings, desktop handoff for /app on macOS and Windows, and Migrate to Codex flows that import setups from Claude Code and Claude Cowork. The migration import is a pragmatic nod to the reality that teams are evaluating across platforms.

What to Watch

  • The public S-1 filing (likely August) will be the most transparent look yet at OpenAI’s financials. Watch for gross margin disclosures and compute cost breakdowns โ€” those numbers will ripple through every enterprise negotiation.
  • Sites pricing is free during preview, but OpenAI has confirmed pricing is coming. The cost model for hosted app deployments will determine whether Sites becomes a serious internal tools platform or remains a prototyping novelty.
  • The twin IPO race: Anthropic filed its S-1 shortly before OpenAI, and both are targeting September-October windows. Two frontier AI labs going public in the same quarter creates a unique benchmark for the sector.

Codex Weekly is a regular analysis for CTOs, engineering leads, and technical decision-makers navigating the rapidly evolving AI tooling landscape. Always consult official OpenAI documentation for current pricing, availability, and deprecation timelines.