OpenAI Codex Sites & Plugins for Knowledge Workers 2026

TL;DR

OpenAI Codex now serves 1 million knowledge workers weekly (20% of its 5M user base), prompting new features beyond coding
Sites lets users build and share interactive dashboards, apps, and custom workspaces via URL with anyone in their organization
Annotations expanded from code to documents, spreadsheets, and presentations — point Codex to specific sections for contextual edits
Vertical-specific plugins launched for sales, data analytics, product design, and investment banking — with finance, legal, and marketing coming soon

What Happened

OpenAI announced three major expansions to Codex on Tuesday, explicitly targeting the knowledge worker market that represents 20% of its weekly active users — roughly 1 million people who aren’t writing code.

The centerpiece is Sites, a feature that transforms Codex from a chatbot into a workspace builder. Users can now create interactive websites, custom dashboards, and shareable apps without leaving the platform. A financial analyst could build a scenario planner that colleagues access via URL. A product team could host launch materials with live updates on messaging and milestones. These aren’t static documents — they’re functional tools that update dynamically.

OpenAI is partnering with Wix, Replit, Figma, Lovable, Base44, and Emergent to enhance Sites with design, deployment, and collaboration capabilities. Business and Enterprise tier customers get access first.

Why It Matters

This is OpenAI’s clearest signal yet that the AI assistant market extends far beyond developers and chatbot users. The company is chasing the same enterprise productivity territory Anthropic staked out with Claude Cowork — and the timing isn’t coincidental.

Knowledge workers represent a larger addressable market than developers, and they work primarily in documents, spreadsheets, and presentations rather than code editors. By keeping these users inside Codex for content creation, editing, and distribution, OpenAI reduces friction and increases lock-in. A sales rep who can prep for customer meetings, pull Salesforce data, and build a shareable brief in one interface has less reason to context-switch to competing tools.

The shift also reframes what “workplace AI” means. Instead of positioning AI as a question-answering layer bolted onto existing software, OpenAI envisions Codex as the primary workspace where AI, human input, and third-party integrations converge. If successful, this makes Codex a platform — not just a feature.

Key Details

Sites

Create interactive websites, dashboards, and apps within Codex
Share via URL with workspace members
Available to Business and Enterprise customers
Integration partners: Wix, Replit, Figma, Lovable, Base44, Emergent
Use cases: scenario planners, product launch hubs, custom reporting tools

Annotations (Expanded)

Point Codex to specific sections of documents, slideshows, spreadsheets, or Sites
AI makes targeted edits or uses selected content as context
Previously available only for code and Markdown files
Now supports standard knowledge worker file formats

Plugins (New Vertical Bundles)

Available Now:

Data Analytics — Snowflake, Databricks, Hex, Tableau integrations
Sales — Salesforce, HubSpot, Slack data for meeting prep
Product Design — design workflow tools
Investment Banking — deal preparation and analysis

Coming Soon:

Corporate Finance
Private Equity
Marketing Strategy
Strategy Consulting
Legal

Future: OpenAI plans to let partners publish plugins directly to Codex and ChatGPT.

Implications

The vertical plugin strategy mirrors Anthropic’s approach with Claude Cowork almost exactly — down to the specific industries targeted. This convergence reveals where both companies see low-hanging fruit: document-heavy workflows where context matters more than task execution.

Sales teams prepping for calls, investment bankers building pitch decks, legal teams reviewing contracts — these are workflows AI can augment today without requiring perfect autonomy. Users remain in control, but AI handles synthesis, drafting, and data retrieval.

The Sites feature, however, is more ambitious. It positions Codex as infrastructure rather than tooling. If teams start building their operational dashboards in Codex instead of Notion, Airtable, or internal tools, OpenAI gains distribution leverage it currently lacks. The question is whether knowledge workers will trust Codex as both creation environment and delivery platform — or whether Sites becomes another underused feature in an already crowded app.

OpenAI’s partner strategy deserves scrutiny too. Integrating Wix and Replit gives Sites technical depth, but it also introduces dependencies. If those partnerships sour or competitors build tighter integrations elsewhere, Sites could feel half-baked compared to native solutions.

Our Take

This is OpenAI playing catch-up, not leading. Anthropic announced Cowork with vertical plugins months ago, and Notion, Airtable, and Coda have been building collaborative AI workspaces for over a year. What OpenAI brings to the fight is scale and brand recognition — 5 million weekly Codex users gives them distribution Anthropic can’t match yet.

But distribution only matters if the product delivers. Sites needs to be more than “Codex can make a webpage now.” The value proposition is shareable, interactive tools that traditional documents can’t replicate — dynamic dashboards, live data connectors, collaborative scenario modeling. If Sites just produces static HTML pages, it’s a gimmick.

The real test is adoption among that 1 million knowledge worker base. Are they asking for Sites, or is this OpenAI pushing a feature it thinks they need? The plugin vertical list suggests OpenAI is guessing (educated guesses, but still guesses) rather than responding to user demand.

Watch for: Enterprise customer testimonials in the next 60 days. If OpenAI doesn’t showcase real companies using Sites for actual work — not demo videos — it means uptake is slow. Also watch whether Anthropic announces similar features. This space is moving fast, and being first to market matters less than being first to product-market fit.