Descriptions:
Craig Hewitt walks through three significant updates to OpenAI’s Codex platform — plugins, sites, and annotations — positioning them as a major step toward Codex becoming a general-purpose knowledge work environment rather than a coding-only tool. The plugins update introduces curated, first-party skill bundles for specific professional domains: data analytics, creative production, product design, sales, and investing. These pre-configured tool sets connect to external services like HubSpot via MCP, replacing ad-hoc prompt engineering with vetted, domain-specific workflows. Hewitt notes that upcoming additions — legal, marketing strategy, strategy consulting, corporate equity — directly mirror the knowledge-work tooling Anthropic has been building, framing it as a competitive land grab between the two AI labs.
The sites feature generates fully functional web pages from a single natural-language prompt. Hewitt demonstrates live, prompting Codex to build a vegan bakery landing page complete with a generated name (Sprout & Crumb), custom graphic, and coming-soon layout — all without explicit design instructions. He argues that web hosting deployment from OpenAI or Anthropic is likely imminent, which would make Codex a full end-to-end publishing platform.
Annotations, the third update, allow users to flag and reference specific outputs within the Codex workspace, improving iterative workflows. Hewitt’s overarching thesis is that Codex is converging on an “everything app” for knowledge workers — a single interface where data analysis, content production, product management, and sales operations can all be orchestrated, with MCP serving as the connective tissue to existing business tools.
📺 Source: Craig Hewitt · Published June 03, 2026
🏷️ Format: Review
