Context Window 62
This week’s stories play out at a distance—in financial markets, academia and big tech—but they point clearly toward future directions for publishers. At the other end of the scale, I’ve written about my hands-on experience with Claude Code and agentic AI, which you could start using today.
The big news this week was financial markets wiping almost $300 billion from the value of data and information companies following Anthropic’s release of a tool for automating aspects of legal research. This might seem distant from some publishers, though big losers like RELX are embedded across the publishing ecosystem, from legal and professional publishing (LexisNexis) to academic platforms (ScienceDirect and Scopus) and exhibitions (London Book Fair). The top end of the sector catching a cold is rarely a good thing overall.
My friend and subscriber Alex Boden had what I thought was the best take on this in his newsletter, highlighting the short-term impact on generic information businesses, but the long-term value of proprietary, high-quality data. Read more broadly, it’s a reminder that publishers’ long-term defensibility lies less in volume and more in originality, rights control and the institutional trust wrapped around their content.
Last week, I wrote about OpenAI’s move into academic publishing workflows. In a complementary development, a team of researchers at Peking University, working with Google Cloud, announced a new multi-agent framework called PaperBanana for generating publication-ready illustrations, diagrams and statistical plots. For scientific publishers particularly, this looks like a really interesting and potentially valuable development.
In parallel, Nature has published an updated policy on AI use which is worth a look as an exemplar for other publishers. There is a lot that I liked: transparency about LLM use in the methods section, differentiating between writing and editing, and most importantly, this beautifully clear statement: “an attribution of authorship carries with it accountability for the work, which cannot be effectively applied to LLMs.”
But, thinking about PaperBanana, it’s striking that the Nature policy is much stricter around images than writing: it refers to unresolved legal issues, but the same surely also applies to text generation? There’s also no reference to use of AI for data analysis, which feels like a notable omission. These are good questions to consider if you’re framing a policy for your own publisher.
More signs this week of walled gardens being built around the web. News publishers are blocking the Internet Archive from their sites, for fear that the Archive will in turn be used as training data by AI companies. And there are reports of Google Books disabling search within book previews, again linked to concerns around training data.
Microsoft launched its Publisher Content Marketplace this week, which not only acts as a pipeline for content into Microsoft Copilot and other applications, but is intended to work with other AI models in future. There’s an impressive list of partners from news, magazine and online publishing, whose input was used to shape the product—but no book or journal publishers listed, so it remains to be seen whether the proposed terms work as well for our corner of publishing.
For educational publishers, it’s interesting to see Google integrating standardised test prep (initially for the SATs) into Gemini, working with educational partners. For some publishers, this points to a partnership opportunity, for others, a new competitor in a valuable market.
I’ve been using Claude Code and Cowork intensively for several weeks, and of all of the AI tools I’ve tried in the last couple of years, they have had the greatest impact on my productivity. This week I helped one of my most thoughtful and creative friends, an artist who is as far from AI boosterism as it is possible to get, build something with Code. His last message to me read, “today actually blew my mind.” Cowork is currently only available for the macOS desktop app, but if you’re a Mac user, the £15 monthly subscription is pennies next to the value it can create (I have no link to or incentive to promote Anthropic products; other agentic frameworks are available).
If you want to know more, I wrote a couple of short essays for my website, first on my experience of prototyping and developing with Claude Code, and then on what the availability of AI agents means for developers and other knowledge workers. I’d welcome your comments on them.
It’s the start of conference season: I’ll be at the IPG Spring Conference in London next week, previewing updates to our training for 2026, and giving a summary of the AI and policy landscape. I hope to see many of you there.
This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.