Context Window 86
To quote a classic film: show me the money… This week, the approval of the Bartz v. Anthropic settlement made that a promise for many publishers and authors. But alongside the legal drama and second-order implications, there’s also a new calculator and a practical guide to agents that I genuinely think everyone should read.
The standout news this week was the approval of the settlement in Bartz v. Anthropic, pending since a fairness hearing in May. The agreed legal fees were reduced again, to a mere $101 million, putting roughly $195 more per author into the settlement pot, depending on other expenses. An extended payment process will now start.
The approval of the settlement has focused authors’ attention on the money that is due to them, but for some, there was an unpleasant surprise when they discovered that some publishers had not registered works for US copyright. As a reminder, registration is not required for copyright protection, but it is required to seek statutory damages in US federal court, making it particularly important for international publishers. Kudos to Macmillan for addressing this head-on: author communications shared online showed the company owning its failure to register some works and promising to make authors whole for any loss. This sets a precedent for other publishers, and it is a salutary reminder to all authors and agents to ensure their publisher has fulfilled any contractual obligation to register works.
Litigation from authors and publishers is leading to a new legal risk for AI platforms and executive decision-makers: institutional shareholders suing Microsoft and Adobe for exposing them to legal jeopardy and reduced valuations. If this encourages executives and boards to conclude that licensing is cheaper than litigation, then more power to the shareholders.
On a more practical note, Aboard published a new guide to agents in the workplace and it is a must-read: expert advice, beautifully written and presented. Page 48 boils the whole thing down to three simple pieces of advice. Print that page out and stick it next to your screen. If you only read one link from the newsletter this year, please make it this one.
Several people shared this AI Environmental Impact Calculator with me: it’s a really useful way of looking at the impact of personal AI use and putting it in context. The methodology is careful to acknowledge that this doesn’t address systemic impact (think data centre builds), and of course, a lot of small incremental impact can add up to something significant. But if nothing else, it’s a useful thought starter.
Creative Access published new research this week looking at publisher use of AI. I had some reservations about the way the research was presented in the trade press and wrote about it on LinkedIn—do weigh in there if you have a perspective on this.
For context on AI use in the broader economy, Gallup published new data showing AI use rising sharply in the second quarter of 2026: nearly half of employees surveyed said their organisation had adopted AI, up 6% in the quarter. Writing and editing is the number one use case, identified by 51% of users. And more than three quarters of employees who had used AI for automation reported increased productivity as a result. Relate this to the previous item: if every other industry is accelerating AI adoption, why would publishing be the exception?
Substack announced that it is integrating Pangram’s AI detection tool for authors and readers. The explanatory post is more nuanced than the headline announcement, concluding that human versus AI writing matters less than aligning author and reader expectations. That nuance matters with tools that are directional, not determinative.
One of my favourite AI tools is being rebranded: NotebookLM is now Gemini Notebook. The alignment with Google’s main consumer AI model (and with other tools like Workspace Studio for automation and agent building) indicates that it’s moving from being a side project to a core part of Google’s AI offer.
One of the most persistent and credible complaints about AI is that it creates debt: technical debt in development, strategic debt in processes, or cognitive debt in individual working practices. An HBR piece this week proposed a new twist on this: brand debt—or the disconnect between what is presented and what is experienced. This strikes me as a clever observation and especially relevant in an industry like publishing, where credibility is one of the greatest assets.
Finally this week, building on an idea I spoke about at the last IPG Lunch and Learn session two weeks ago, I wrote a new essay about books, AI training and a historical precedent from shipwrecks and salvage. The Venn diagram of people who share my enthusiasm for AI, publishing and niche historical parallels is tight—if that’s you, you’re in an exclusive club 😉—but I hope it might give you a different perspective for the weekend.