Context Window 64

In recent months there has been a discernible shift in AIand the pieces I featureaway from simple chatbots to agentic tools like OpenAI’s Codex and Claude Code. My smartest friends have moved from using AI for five or ten-minute microefficiencies to delegating complex coding or research tasks measured in hours. Chatbots will still have their place, but the rise of AI agents and big themes around trust dominate my thinking this week.

I’ve been in Madrid this week for Parix Audio Day, where I gave a keynote address on the impact of AI on audiobook publishing, including content creation, narration and the competitive dynamics that arise when audiobook production and distribution are bound up in the same big tech platforms. My slides from the day are available on my website.

Andrew Savikas (who OG digital publishers will remember from O’Reilly’s Tools of Change for Publishing conferences) wrote a great piece on determining the future of knowledge work by watching how developers use AI agents. It resonated with a lot of my recent experience and thinking.

I wrote a piece in response about how developers and publishers respectively use AI, the skills publishers bring to the table, and what each group can learn from the other.

Technical editor Rachel Andrew published an excellent piece this week on how AI has broken the implicit contract between subject-matter experts and their editors. It raises really important questions about trust and authenticity.

I hadn’t planned to write a second long-form piece this week, but on the flight to Spain I wrote some thoughts on what Rachel’s piece means for book publishers and how to build trust in the author-editor relationship. I’d really welcome comments on that, because it feels like the second and third-order effects of a broken author-editor contract are serious.

Related to this, thanks to subscriber Avi Staiman for sharing his recent articles for Scholarly Kitchen, on why journal authors aren’t disclosing AI use and what publishers should do about it. I particularly liked his insight that many publishers are treating AI use as a distinct, trackable event, but that distinction blurs as AI is integrated into day-to-day tools. That insight applies universally, across every kind of publishing.

And on that note about integration into tools… Ian Mulvany of the BMJ wrote an excellent piece responding to OpenAI’s Prism tool for authoring scientific research. He argues that the real impact of AI isn’t any single tool, but how it reshapes the environments where people actually work. For publishers, that means thinking less about hosting content and more about ensuring it flows into—and adds value inside—those new AI-driven workflows.

The shift to agentic AI is also reflected in the mainstream media, with two good pieces if you want something more expansive to read over the weekend: for The Atlantic, Lila Shroff argues that we’re now in a post-chatbot era, and in the New York Times Opinion section, Paul Ford takes a nuanced, careful look at what agentic tools mean for software, employment, and society. I love this quote:

I believe there are millions, maybe billions, of software products that don’t exist but should: dashboards, reports, apps, project trackers and countless others. People want these things to do their jobs, or to help others, but they can’t find the budget.

To that point, my friend Matt Webb just released a collaborative, online editor for Markdown files. That might be relevant to some subscribers, but the broader point is not so much the what but the how: it was created by spending half an hour dictating a specification into his Apple Watch, and then using the transcript as a prompt for Claude Code. It’s amazing to watch previous frictions in software development drop away: in this case, if you can articulate it verbally, you can prototype it. What niche application or workflow could you describe, and what’s stopping you from trying to build it?

If you do try building something, and you want it to look professional, design platform Figma released a new integration with Claude. It allows a user to create an app prototype in Claude Code and then develop the interfaces and user experience within Figma.

For anyone thinking about the visibility of their books on the web, this post from Ahrefs summarises multiple research projects on over a billion data points on AI content and citations. Key takeaways for publishers include the importance of particular media types for visibility: YouTube videos, Wikipedia, listicles, all of which should form part of a rounded content strategy.

This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.

Written on February 20, 2026