返回研究新聞
An East Asian educator points to a reviewable project timeline on a desktop while a Black adult learner holds the same supervised workflow on a phone
政策 / 倫理產業訊號20262026年7月26日· 13 min

Product news: ChatGPT Work, Claude Cowork and Gemini extend guided work across devices and classrooms

500 字摘要

High-angle view of a Black adult learner reviewing a phone approval between a desktop workstation and an East Asian educator with a South Asian learner checking printed evidence; a closed folder remains off the cyan route.

This product-news report follows a shift that is easy to overlook when attention stays on model benchmarks: AI work is becoming portable, persistent, and connected to the places where people learn. OpenAI's current ChatGPT desktop update lets cloud Work conversations and Project context move across web, mobile, and desktop. Anthropic is extending Claude Cowork from the desktop to web and mobile while delegated work continues in the background and consequential decisions return to the user. Google is linking adaptive study notebooks in Gemini with NotebookLM and teacher-led activities in Google Classroom. Together, these products make continuity across devices and institutional contexts a central design question for AI in education.

OpenAI's July 16 release notes describe a clearer desktop separation between Chat for conversational help and Work for longer, end-to-end tasks. Chat and Work conversations now share a unified recent-items view, existing ChatGPT Projects appear in the desktop app, and users can begin a cloud Work conversation on one surface and continue it on web, mobile, or desktop. Local Work conversations remain on the computer. The distinction is important for schools and universities: a project can preserve approved files, prior instructions, and work history across settings, but local and cloud tasks do not have the same data boundary.

For teaching and academic work, that continuity could support a research project that begins in a seminar, continues during independent study, and returns for review without repeatedly rebuilding context. Faculty and staff could also carry a multi-step analysis or drafting workflow between office and campus devices. Yet convenience can obscure what the system retained, which files shaped the result, and whether a mobile review was substantial. Institutions should label cloud and local work clearly, define which project materials may sync, and require process evidence when the task is assessed.

Anthropic's July 7 Claude Cowork announcement presents a similar transition. Cowork can operate across connected files, calendars, email, messaging, the web, and other tools until a delegated task is complete. The new web and mobile rollout, beginning in beta with Max users and expanding to more plans, allows a person to start work at a desk, monitor or redirect it from a phone, and retrieve the result elsewhere. Background and scheduled tasks can continue when the original device is offline. When the system reaches a decision reserved for the user, it sends the question to the phone; Anthropic says the output remains subject to human review and approval.

This design could help a teacher monitor a long-running resource search or let a project team review a draft between meetings. It also makes oversight more demanding. A notification is not proof that a person understood the preceding steps, and approval on a small screen can encourage superficial checking. The desktop retains the fuller Cowork experience, including local files and browser use, so capability and risk differ by surface. Educational pilots should identify which tools and records are available on each device, preserve an auditable plan and action history, and reserve enough time and screen space for meaningful review.

Google's June education release provides the most explicitly pedagogical version of portable continuity. Study notebooks in the Gemini app begin with a learner's goal or uploaded course materials and a diagnostic quiz. Gemini then creates bite-sized lessons and updates the plan as quiz results and new materials arrive. The notebooks can connect with NotebookLM for source-grounded questions, flashcards, and other study artifacts. Google also announced no-cost ACT and GRE practice tests, with performance broken down by topic to guide subsequent study.

The institutional bridge is Google Classroom. Google says students of all ages will be able to use Gemini in Classroom for study guides, quizzes, and Guided Learning grounded in course materials. Teachers will be able to assign a study notebook from Classroom and receive insights about where an individual learner or class may need support. The release is staged: study notebooks launched first for adult personal accounts, were scheduled to reach school-issued accounts later, and were desktop-only at announcement, with mobile support planned for later in the summer. Those limits matter when institutions evaluate access and equity.

None of these announcements establishes that cross-device persistence improves learning. Continuity may reduce setup friction, but it may also weaken stopping points where learners normally recall, reorganize, and explain what they know. Background agents can complete work while the learner is absent; adaptive notebooks can personalize activity while concealing why the pathway changed. Product evaluation should therefore measure independent understanding, not only completion speed, engagement, or the smoothness of handoff.

For Hong Kong schools and universities, the practical response is to design continuity with boundaries. A pilot should map what context follows the user, distinguish local from cloud storage, restrict connectors, show students when an agent continues without them, and require review on an appropriate device before anything consequential is submitted. Teachers should be able to inspect sources, intermediate actions, diagnostic evidence, and changes to an adaptive plan. Portability can make support more available, but educational value depends on whether learners remain able to pause, question, reconstruct, and take responsibility for the work.

相關論文

Eight diverse adult educators work in three small groups with generic text-free laptops while a mentor guides a practical workshop discussion
政策 / 倫理2026年7月30日
政策 / 倫理 49

OpenAI product news: AI Skills Jam brings hands-on AI practice to K-12 educators

OpenAI

AI Product and Learning Report

Product news: OpenAI Academy and the Walton Family Foundation announced hands-on AI Skills Jam workshops for more than 1,600 US K-12 educators and leaders, linking practical experimentation with continuing resources while leaving learning impact to be independently evaluated.

product newsteacher professional learningAI Skills Jam
閱讀 500 字摘要
A Black IT lead, East Asian educator, and White governance officer inspect three geometric system modules linked through permission gates and a human approval control
政策 / 倫理2026年7月29日
政策 / 倫理 47

Anthropic product news: MCP 2026-07-28 brings stateless, governed connectors to Claude

Anthropic, Model Context Protocol

AI Product and Learning Report

Product news: Anthropic says the July 28, 2026 Model Context Protocol specification introduces a stateless core, versioned Apps and Tasks extensions, and stronger OAuth-based authorization; schools still need least-privilege access, human approval and auditable data governance.

product newsMCP 2026-07-28institutional AI governance
閱讀 500 字摘要
A Black teacher and East Asian and White secondary students compare image and video frames across a monitor and tablet with restrained cyan provenance markers
政策 / 倫理2026年7月28日
政策 / 倫理 45

Gemini product news: SynthID verification gives schools a media-provenance check, not a truth detector

Google for Education, Google DeepMind

AI Product and Learning Report

Product news: Google Workspace for Education directs learners and educators to upload media to Gemini and ask whether SynthID identifies Google AI generation or editing; the signal can support provenance lessons but cannot establish accuracy, intent or human authorship when absent.

product newsGemini media verificationAI literacy
閱讀 500 字摘要