Context Window 82
There are twin themes this week: administrative convenience in tools you are already using like Slack and Excel, and more ways for creators to control their output—written, spoken or performed. And for publishers, two important new research projects. If you do one thing this week, please take ten minutes to respond to whichever is relevant to you—it’s how we move the industry from anecdotes to data.
Anthropic released a new feature called Claude Tag, which adds Claude to Slack teams—workers just need to mention @Claude in a conversation to be able to delegate tasks to it. Importantly, it will also have access to previous chats and files for context.
It sounds extremely convenient and the sort of thing I would dive into head first, though from a practical perspective I note that it uses the token-hungry Opus 4.8 as its base model. So I wonder if there will be a tension between colleagues using it and administrators worrying about cost control…
Of particular interest to anyone who came to my course on data analysis last week or on agentic AI for Microsoft platforms this week, Microsoft announced a significant upgrade to Copilot for Excel yesterday, including third-party data sources and the ability to create repeatable Skills—potentially very applicable for operations and finance teams.
This is a really useful step: there are definite use cases for AI-voiced files, but in the last year I’ve also seen numerous cases of unscrupulous publishers listing AI-generated audiobooks with human narrators, doubtless to get around Audible policy requirements. It’s also timely, with the transparency obligations in Article 50 of the EU AI Act applying from 2 August 2026.
It has been developed by a non-profit co-founded by the actor, and is an extension of the existing Really Simple Licensing standard for web publishers. For publishers and agents, I can imagine a lot of prominent authors being as interested in this as Blanchett’s fellow actors.
For smaller publishers or individual creators looking to have more control over how AI models use online content, Cloudflare launched a new partnership this week with beehiiv which will see the newsletter provider integrate Cloudflare content controls into its platform, a significant step up from what rival platforms like Substack offer.
I’m a huge fan of email as a core marketing channel and with this development, a newsletter platform is no longer just a CMS and mailing tool. It may determine what data and leverage a creator has over AI access.
The Book Industry Study Group and BookNet Canada have launched their 2026 AI in the Book Industry Survey, open to anyone working in the North American book trade. Similarly, on the other side of the water, the Independent Publishers Guild is running a survey of its members in partnership with Oxford Brookes (the link is on your last IPG member bulletin). Getty announced a multi-year agreement with OpenAI, which will see licensed images appearing in ChatGPT search and outputs—though important to stress that this is about discovery, not model training.
Also on licensing and monetisation, congratulations to Publishers’ Licensing Services, which announced that over 250 publishers have now signed up to its new collective AI licence. I’ll be at the PLS Conference next week and will write up any AI-related reflections for the next newsletter.
The Independent interviewed three authors on how they use AI in their workflows: tl;dr, research rather than writing, but the case study of how business journalist Katie Prescott used NotebookLM to research her biography of late Autonomy founder Mike Lynch is pretty awesome.