Context Window 102
It’s been a turbulent week in a turbulent year. The scandal enveloping Nvidia has rocked financial markets and tech stocks, and as I’m preparing to send this on Friday morning, Microsoft is experiencing the second day of one of its worst outages for years. The implications of the former may take time to play out. I suspect the latter will be more immediate and that I’ll see it in today’s newsletter metrics. In the meantime, if you’re struggling to access Microsoft systems today, the newsletter has some practical things you can try, from setting up a ChatGPT Business Profile to testing new features from Claude and Adobe.
OpenAI launched a set of business tools this week. This includes the ability to claim a business profile on ChatGPT by providing details of your company, including a product metadata feed, and verifying ownership through a code or custom DNS record (though so far, it’s timed out every time I’ve tried this for my own business).
In addition to shaping information about their company being surfaced in ChatGPT outputs, business profile owners get access to an analytics platform with data on how often the business appears in outputs, and website clicks. There’s also an option to create business agents, though this is currently restricted to US e-commerce and online service businesses only. However, the data alone makes this a worthwhile step for any publisher thinking about GEO—which should be everyone.
Anthropic also shipped new features, including the development of its dashboard artifact (which I wrote about here) into Claude Data, and the ability to run third-party apps within Claude sessions.
However, the one that interests me most is Claude Whiteboards, a collaboration feature for teams that allows collective brainstorming, organisation of ideas and functionality to synthesise that thinking into a document. There’s also a new Claude Meetings feature, an agent that can join online calls, take notes and organise information. There’s a short product demo here which is worth a minute of your time. At face value, it looks extremely interesting for remote teams—I’m facilitating an online workshop next week and will be giving it a go.
For creative and design teams, Adobe’s Max 2026 event saw a range of AI features released. The Firefly app gets an AI workflow feature, Firefly Graph, and there is a new AI assistant being added to InDesign, Illustrator and other tools. At the same time, Adobe previewed MCP integrations to be released next year, which allow creative workflows to access information stored elsewhere—a design brief saved in Google Drive, for example—and to be controlled by other AI tools. The dual strategy of enabling AI tools within their software and supporting third-party models recognises that creative teams and individuals will build AI into their workflows in different ways.
Also on the visual front, Instagram introduced a new Effect Builder tool, which allows users to describe a filter, effect or aesthetic with a text prompt and apply it to Instagram Stories—useful for social media content creation. Like OpenAI’s business agent, it’s US-only at launch.
Audible announced an expansion of its consumer AI features, including story recaps and asking questions within the audiobook. The features were piloted on public domain and Audible original titles and are now being introduced to a wider list of titles, ahead of a broader rollout to English-language titles in December. Credit to Audible for a reasonable level of clarity and control: the announcement specifies that audiobook content is not retained or used for training by their AI. ACX authors and publishers have the opportunity to opt out of the features if they choose. It’s not clear what arrangements are in place for publishers working with Audible on traditional contracts, but it’s a good prompt to pick this up with your Audible contact if you have not already heard from them.
Moving from the product side to how AI is being used, the Caine Prize for African Writing has confirmed that its 2026 award stands, after a piece in The Atlantic alleged AI usage based in large part on Pangram data. Helpfully, the prize breaks down its investigative process, its findings, and the new AI policy developed as a result of this episode—useful for anyone reckoning with provenance questions. It’s evident that what cleared the winner was analysis of metadata from four Word documents: writers, save your drafts. It concludes with a useful reminder:
“AI detection tools alone are not a reliable basis for determining authorship.”
Victoria Strauss published an update on the Anthropic settlement process: written from an author perspective, it will be useful to anyone who made a claim and discusses the process of resolving disputed claims and the importance of registering US copyright for titles.
Over 100 scholarly publishers signed a new university press charter, which includes undertakings on responsible AI use (disclosure: Bristol University Press, where I serve on the advisory board, was involved in developing the charter). It’s a useful starting point, though for me the key question is how these good intentions are translated into granular policies that inform day-to-day decisions for publishers and authors. As we saw with the AI charter recently introduced and withdrawn by the Society of Authors, there is a broad range of opinion on what exactly constitutes “responsible” use.
On the subject of education, the UK’s Office for Students and OpenAI published major new reports on how AI is being used in teaching and learning. The picture they paint is of reading being displaced, with AI summaries replacing engagement with core reading materials, and a gap between usage metrics and learning outcomes. Both will be of interest to academics, educational publishers and edtech companies.
Finally this week, I’ve been plugging my forthcoming book, but for an excellent, highly complementary take, can I recommend AI by Design: The Human Way to Make AI Work by James Lascelles, Rob Maguire and Aslam Jilani, which publishes next week. Written with real wisdom and clarity, it focuses on the most important aspects of AI adoption: people, processes and culture. I really enjoyed it, and if you like this newsletter, I think you will too.