Everything is now a data business, especially media companies. My friend Alex Boden’s analysis of the Washington Post’s pivot to WP Intelligence is characteristically sharp: editorial expertise converted into structured intelligence products, sold to professional audiences on enterprise contracts. The playbook works for a news publisher. The question for trade book publishers is what version of that pivot is available to them.

Thanks to Luis González, Javier Celaya, Christopher Kenneally and their colleagues for inviting me to give a keynote address on the impact of AI on audiobook publishing this morning at Parix Audio Day 2026 in Madrid. It’s one of the best conferences in the publishing calendar, and hands down the best venue.

I’ve been trying to write something here roughly once a week, largely but not exclusively about my experience working with AI in publishing. I hadn’t planned to post anything else this week, but John Willshire linked to a post from Rachel Andrew on managing publishing operations in the context of generative AI and I wanted to respond to it with some thoughts about how the issues it raises affect book publishing.

Andrew Savikas recently wrote that one of the best ways to see the future of knowledge work is to watch how software developers work. It made me smile because I got to know Andrew nearly twenty years ago through his work at O’Reilly Media and their Tools of Change for Publishing conference—“watching the alpha geeks” was a core philosophical and strategic tenet for O’Reilly.

I’m on my way back from the IPG Spring Conference in London, always one of the highlights of the publishing year. It was a really good mix of the inspirational, the practical, and a uniquely generous, welcoming and supportive community. I particularly enjoyed the presentations from former Shazam chairman and crime writer Ajay Chowdhury, National Literacy Trust CEO Jonathan Douglas, BBC media editor Katie Razzall, Illumicrate CEO Daphne Tonge and Bonnier UK Co-CEOs Sarah Benton and Jonathan Perdoni. I spoke in the morning about the IPG’s forthcoming short training courses (data, content creation, websites/GEO and AI risk/governance) and in the afternoon to give members an update on the policy and licensing landscape for AI.

I had a couple of conversations this week that inspired a quick coda to my previous post on using Claude Code for prototyping ideas. The day after I posted it, I spoke with Tom, one of my most thoughtful and creative friends—someone who is about as far away from AI boosterism as it would be possible to get. A decade ago, he had spent several thousand pounds with a web developer building a location-aware mobile website. He asked how much information Claude Code needed to build something similar. I copied his sixty-word WhatsApp, clarified half a dozen questions that Claude asked in response, and it took under two minutes to replicate the core functionality.

Last week, Ethan Mollick posted about creating a functioning game with Claude Code and a one-shot prompt. As a first project with Claude Code, it seemed ambitious but appealing. I started with a single paragraph prompt to develop a short, simple adventure game in the style of the Sierra On-Line games I remember from the late eighties, with an AD&D 2nd Edition vibe. Getting a first, testable version was almost indecently fast. It then took several rounds of iteration to improve it, including a separate workstream to create retro style screens (Claude doesn’t have native image generation). The total active investment of time was about twenty minutes, with Claude Code running in the background while I did other things.

There’s an uncomfortable paradox at the heart of Generative AI adoption: the promise from AI vendors and consultants like me is that the technology will boost productivity. And for the most part, this holds true with my clients. But the gains are unevenly distributed. Dan Hon wrote in his newsletter recently that his friends who use LLMs brilliantly in their coding work do so because they bring very particular skills and experience that don’t translate to the average user—a challenge amplified by what Ethan Mollick calls AI’s ‘jagged frontier’ of uneven capabilities. Scale this up to enterprise level and you hit a bigger problem: AI can create efficiencies for individual, power users, but it doesn’t necessarily deliver the company-wide transformation the C-suite was sold on. You’ve probably seen headlines claiming that most AI pilots fail to deliver measurable P&L impact. Are the promised gains genuinely elusive, or do we need to reframe what we’re looking for?