I’ve been on the road again this week, travelling to Portland, Oregon for the Independent Book Publishers Association Publishing University, where I’m speaking today. One quick observation, informed by email exchanges with my friend and his AI agent while I was travelling, and several conversations at the convention.

Earlier today I attended an online seminar hosted by the Creative Industries Policy and Evidence Centre, where Dr Paul Crosby from Macquarie University presented recent research on authors and AI carried out in Australia. Many of the sentiments expressed in the research were familiar: author concerns about the morality of AI training and the impact of AI outputs on their livelihood echoed many of the findings of the Cambridge research on AI and the novel published last November. What I found most interesting about the Australian research is that it was interpreted through economics and econometric analysis. One of the key questions was whether, if AI companies train on human creative work without compensation, the long-run incentives for cultural production are weakened. The concern was that more economically marginal forms of publishing such as literary, niche and experimental writing could be hardest hit. Framing the impact on authors and the creative industries as an externality is a helpful way of emphasising to policy makers that the economic benefits of training on copyrighted work accrue overwhelmingly to tech companies in the US, while the costs are potentially distributed across the wider cultural ecosystem globally. This is a textbook setup for undersupply of a public good (creative works).

I’ve been leading training sessions on generative AI for nearly three years now, and over that time the advice that I’ve given on prompting has been relatively consistent: good results generally depend on a degree of detail and specificity about context, objective, and outputs. I was interested to see Ethan Mollick noting a change of emphasis in the developer documentation for GPT-5.5, OpenAI’s latest thinking model:

I’m writing this on the way home from Chicago, where I gave a keynote on AI and publishing at the ECPA Leadership Summit. As I said in my presentation, this is not an area of publishing I’ve worked in, and my role was to provide an outside perspective—less about the specifics of the sector, more about what AI might look like when viewed from beyond it. It was a really rewarding and thought-provoking experience, and I am very grateful to Jeff Crosby and his colleagues at the ECPA for their hospitality.

It has been a busy week: updating the materials for and delivering a lunch-and-learn session and another full day training course on AI for IPG members, and preparing a keynote presentation on the future of publishing for a conference in Chicago next week. To prepare for those things, I’ve been reading and talking to a lot of interesting people about how they are using AI at the moment. In between, I made a quick dash up to London for my ten year business school reunion—a room full of smart people from different industries who provide a triangulation to the publishing experience. There are three interesting patterns that kept coming up in all of those different contexts.

If you read my piece last year on sizing and plotting the UK publishing industry based on Companies House data, there’s an interesting coda today with the publication of the Office for National Statistics’ updated Standard Industrial Classification (SIC) codes for 2026. If you’re not familiar with what SIC codes are, you’ve probably given up reading by now—but the short version is that SIC codes are short numbers that describe what business activities companies engage in. Firms choose one or more SIC codes when they register with Companies House.

Ian Mulvany has a good piece in the latest issue of InPublishing examining challenges and opportunities for publishers. A couple of his points really stuck with me. Under pressure from investors to signal progress, companies are forcing AI into products and services without a clear sense of value. The result is a proliferation of new features that are individually justifiable but risk being collectively incoherent. And while generative AI can speed up the creation of content, code or even products, it doesn’t create additional time—in other words, opportunity cost still applies. I’ve started thinking of this pattern as strategic debt.

I took part in a webinar on AI and publishing earlier today, hosted by the Crius Group, alongside my friends and colleagues Cameron Drew and Simon Mellins. One of the topics that came up was what mistakes we saw organisations making with AI. The conversation was about publishing, but the more we talked, the clearer it became that these failure modes aren’t industry-specific. They show up everywhere. Here are eight of the most common. If you recognise more than two or three of them in your organisation, you may not have an AI problem—you may have a strategy and operating model problem.