Product news: ChatGPT plugin ranking and Claude Code 2.1.239 make tool selection and workspace boundaries inspectable

This product-news report connects three approaches to choosing and governing AI tools. OpenAI's August 21 ChatGPT update changes plugin discovery and several context-facing parts of the product. Anthropic's Claude Code 2.1.239 makes cloud-synced plugins and some infrastructure costs more visible. Google positions Gemini for Education around teaching, learning and institutional work. The shared educational question is whether users can see why a tool was selected, which workspace supplied it, what context shaped the result and who remains accountable for checking it.

OpenAI says plugin recommendations on ChatGPT web and mobile now prioritize tools that people continue using after installation. Availability still depends on plan, region and workspace settings, and desktop is outside this update. ChatGPT also gained more awareness of a user's local time, faster loading for long web conversations and progressive display of interactive content. These changes may reduce friction, but popularity or retention is not evidence that a plugin is pedagogically appropriate, privacy-preserving or accurate for a specific course.

In education, recommendation should be treated as a prompt for inspection rather than an endorsement. A teacher or student should identify the plugin owner, requested permissions, connected data, expected output, accessibility, cost and removal path before installation. Time-aware answers can make deadlines or study plans more relevant, but users should still state the authoritative timetable and time zone when consequences matter. Progressive rendering improves responsiveness; it does not mean an unfinished interactive artifact is ready to use.

Claude Code 2.1.239 addresses related boundaries in agentic development. Plugins synchronized from claude.ai are labelled with a synced namespace, can be enabled or disabled explicitly, and do not replace a locally installed plugin with the same name. The release also includes the 1.1-times United States-only-inference premium in cost estimates for data-residency workspaces. Other fixes preserve plan mode across idle cloud-worker restarts and recover remote MCP servers after transient reconnect failures.

These controls improve legibility without proving correct execution. A synced plugin and a local plugin can carry different instructions, capabilities or trust assumptions even when their names resemble each other. A visible cost estimate helps budgeting but does not establish that a data-residency choice satisfies institutional policy. Student teams should record the exact plugin identity, source, version, workspace, permissions, data route, estimated cost, actual artifact and acceptance test.

Gemini for Education supplies the institutional comparison. Google describes educators using it to plan and differentiate lessons and create assessments; learners using it for explanations, practice and writing feedback; and staff using it for communications, administration and research. Google also describes administrator controls and education data that is not human reviewed or used to train AI models. These are vendor-described capabilities and protections, not independent evidence of learning impact.

For AIEDHK, the practical response is a tool-selection ledger. Before an AI extension enters a course or research workflow, record why it was recommended, who approved it, which account and data it can access, where it came from, what it costs and how its output will be verified. Product interfaces are becoming better at showing selection and context boundaries; institutions should turn those signals into explicit, teachable review practice.