Context Window 83
It’s been another busy week: attending British IP Day at Parliament and the Publishers’ Licensing Service conference, and getting ready for three workshops in Dublin, Cambridge and London next week. As a result, I almost missed a milestone: Wednesday marked six full years of my consulting business. I’m very grateful to clients, newsletter subscribers and everyone else who has made those years so interesting.
The most practical thing I read this week was not solely focused on AI: an article on how creative people can stay discoverable when search, social and AI are no longer reliable referral channels. It suggests eight short principles, all of which tally with my experience around newsletters, websites and LLMs.
Much of this advice also applies to publishing brands, but there is a simpler, more universal question for publishers: what advice are you giving your authors and contributors on building a sustainable promotional platform, and is it as clear as this? Based on some of the author networks and conversations I am in, I think there’s a significant opportunity for publishers to provide training and support to authors on platform building in 2026.
Publishers wanting inspiration around visibility and discoverability could look at this DigiDay round up of how leading online publishers are responding to AI crawling: the range includes blocking bots, redirecting them to specially optimised pages which are easier to crawl, and allowing access through WebMCP. If I were running an Open Access publisher, the redirect/optimise approach would look very interesting to me.
Microsoft announced the formation of a new business, Microsoft Frontier Company, with 6,000 engineers who will be embedded at customers to develop AI solutions. It seems to be an admission that however good the model, AI transformation is as much about human judgement as technology.
Meanwhile, more evidence of AI costs starting to bite, with large technology businesses cutting back on employee usage. Most publishers will not face seven-figure monthly AI bills, but uncontrolled usage can still become a real operating cost. The lesson is not panic, but measurement: know who is using which tools, for what tasks, and at what cost before usage becomes habitual.
Google’s Gemini Spark agent is being updated with more app integrations, including Canva and Dropbox. It’s also being rolled out in the Gemini desktop app for macOS. The most useful update for me is remote support, so I could give an instruction on my phone on the move and have it executed on my desktop.
Promises of investment have been part of the government relations playbook for AI companies for some time, but this is the first time I can recall an explicit offer to rights holders being part of the package. The Australian government had previously rejected a TDM exception, and insisted it had no plans to weaken copyright protections, but this will be one to watch carefully.
One of the points made by creatives in the Guardian coverage is that they are ready to do licensing deals, a theme that also came out this week at British IP Day on Tuesday and the PLS Conference on Thursday. PLS now has over 300 publishers signed up to its Generative AI Licensing.
Anthropic released a new app, Claude Science, which brings together a suite of tools aimed at research scientists, including literature analysis, research, generating figures, reviewing manuscripts, and providing an audit trail for findings. As with earlier developments in this space such as OpenAI’s Prism editor, the real advantage for AI companies is moving from training on outputs to potentially training on the upstream process. For scientific publishers, the questions will be around working with the outputs of this kind of environment.
For academic subscribers, the Association of College and Research Librarians has launched a new, three-module online course on the impact of AI on campus, supported by Clarivate. An earlier, introductory course reached nearly 10k professionals.
Finally this week, this viral quiz on attitudes to AI is nicely done, but I also found working through the questions genuinely useful as a reflective exercise. Which of the 30 AI archetypes are you closest to?