Context Window 32
This edition covers Amazon CEO Andy Jassy’s bullish AI update and what billions of agents mean for supplier teams, Turing Institute/LEGO research on how children and teachers are using generative AI, a PLS consultation on licensing for AI training, new image-generation tools from Midjourney and Adobe Firefly, ChatGPT’s new Record mode on Mac, Reddit’s Community Intelligence product, a New Yorker piece on what AI is doing to reading, and the resignation of UK PM AI Adviser Matt Clifford.
It’s been a long week. TGIF. Amazon CEO Andy Jassy gave a bullish update on the company’s use of AI, predicting a world of billions of AI agents and a reduced Amazon workforce. That employment impact led most of the reaction to his update, though I’m more interested in the second order effect on Amazon’s suppliers, who will make up part of the ecosystem of billions of agents that Jassy predicts. Over the course of my career I’ve seen roles, responsibilities and tasks in sales and marketing teams evolve to better service what is, for most publishers, their biggest single customer. It seems clear that many of their future interactions will be with agents rather than Amazon employees. Controversially, I’d suggest that optimally those touch points would be agent-to-agent with human supervision (that is, largely synchronous, real time, always-on interaction). The high-performing sales and marketing team in 2025 has AMS/metadata specialists. In a year or two, I’d suggest it will be built around agent/MCP expertise, freeing up human time to work with non-agentic customers. The Turing Institute, supported by LEGO, published new research on the use of generative AI by children. If you’re a children’s or educational publisher, a parent or just curious about the future, it’s essential reading (the research is based on UK data, but I’d suggest the broad pattern would be seen in other countries). A few highlights: nearly a quarter of children aged 8-12 have used generative AI, largely ChatGPT; more than two thirds of teachers are using AI in their work, and one of the key use cases is developing more personalised lesson plans, with more than 85% saying it had increased their productivity and 88% feeling confident in the tools they are using; however, nearly three quarters of teachers used personal rather than institutional AI accounts, raising significant concerns about security and data exposure. Publishers’ Licensing Services are running an important consultation on licensing published content for Generative AI training. You need to be a PLS-registered publisher and license content through the CLA to respond, but as I hope that includes many of my subscribers, please respond if relevant to you. There were two significant image generation updates this week. Midjourney released its v1 model, which generates short video clips from still images. Potentially very useful for developing imagery for socials/product pages, though after last week’s litigation news, don’t feed it your Disney vacation photos. And Adobe released mobile apps for its Firefly image generator, available on iOS and Android. ChatGPT introduced a Record mode, allowing users of its Mac desktop app to transcribe and summarise audio—for example, recording meetings. This could be a useful productivity tool, but of course there are important questions about transparency, consent and security. I’ve written before about Reddit licensing user data to AI companies and what a significant source of data it has become for LLMs. For marketers looking to exploit the wisdom of that very large crowd, Reddit introduced a new tool this week, Insights/Community Intelligence. This is aimed at major advertisers on the platform, but could be useful for brand marketing. I’ve seen a lot of discussion this week of a New Yorker piece titled What’s Happening to Reading? The essay grapples what we gain and lose as AI transforms our relationship with text. It acknowledges that summarisation and AI can extend access, helping neurodivergent readers or clarifying difficult passages. However, it also raises deeper concerns about whether something essential is lost when text becomes endlessly malleable and personalised. The piece suggests we may be moving toward a world where reading becomes less about sustained encounter with an author’s vision and more about efficient information extraction. This isn’t just nostalgia—it’s about whether we lose shared cultural touchstones and the transformative potential of wrestling with challenging ideas. As publishers, this presents both opportunity and responsibility: how do we create adaptive formats that meet readers where they are while preserving spaces for the kind of deep, sometimes difficult engagement that has historically shaped minds and culture? The question isn’t whether AI-assisted reading is good or bad, but how we navigate this transition thoughtfully. STOP PRESS for UK readers: I normally have this written by Thursday night but as I was doing final checks, I saw the news that the Prime Minister’s AI Adviser Matt Clifford is stepping back from his role.
This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.