Context Window 4
This edition looks at Steven Johnson’s Long Context project, the Activate 2025 outlook on AI search and disintermediation risks for publishers, the rise of AI shopping assistants like Amazon Rufus, and tech news including Amazon’s expanded Anthropic investment and the new Model Context Protocol standard.
I really enjoyed this microsite created by the author and AI researcher Steven Johnson, which includes a fun piece of interactive fiction generated by AI (and the prompt used to create it). More importantly, the article that accompanies it has interesting things to say about memory, context and how books and authors work with AI: ”In a long-context world, maybe the organizations that benefit from AI will not be the ones with the most powerful models, but rather the ones with the most artfully curated contexts.” Activate Consulting’s 2025 Technology and Media Outlook has some interesting insights for publishers on AI and, though it’s not the focus of this newsletter, you might also be interested in their data on US audiobook consumption. Two key takeaways for me were AI search/discovery, and how publishers avoid AI disintermediation. Apposite as I’m sending this on Black Friday: 51% consumers aged 18-34, and 41% aged 35-54, are using generative AI tools for shopping and research (page 44 in the report if you don’t want to browse). Consider this in the context of book discovery. Self-publishing platforms have already driven at least an order of magnitude increase in the number of books published each year, and AI will push that further. Stores arguably have too much choice in certain categories. (My eldest daughter, a b-school student, recently reminded me of the paradox of choice, the idea that reducing consumer options lessens anxiety and drives purchase behaviour). Of course, some book purchasers know exactly what they’re looking for, but in future, will browsers wade through categories and searches, or ask an AI tool to make a recommendation that cuts through the bloat? How do individual books stand out in a marketing context that evolves from optimising for keyword-based search to optimising for AI-based recommendations? And should publishers be developing AI-based recommendation tools for their own websites and channels? It’s a good conversation to have with your CMO after the holidays. Incidentally, if this feels too much like future gazing, consider that Amazon Rufus, an AI shopping assistant, launched last September. At the moment it seems to be trained on metadata and customer reviews, but the difference between books and non-media SKUs is that the product is the training data. Relating it to the Steven Johnson piece, it’s better context. The report also has a good set of insights on how publishers can avoid getting disintermediated by AI search (page 36 this time), which I talked about in the context of the news business a few weeks ago. The recommendations on interactivity and user experience are most relevant to non-fiction publishing, but the point on gating websites is relevant to all publishers. Tech developments this week: Amazon announced a further $4 billion investment in Anthropic, taking its total investment to $8 billion. And Anthropic announced a new, open source standard called Model Context Protocol to allow AI systems to interact with data in business systems and popular cloud repositories like Google Drive and GitHub.
This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.