Context Window 98
The furious reaction from authors to a new charter for ethical AI use introduced by their UK trade union this week demonstrated the strength of feeling about AI in some quarters. But at the same time, lots of thoughtful people are using AI to do interesting things around media and publishing: there’s a strong practical bias this week: new tools you can try for social media and data analysis, and examples of good things people have built with AI. Ethical concerns and practical utility can both be true at once, which is the uncomfortable bit.
It’s been a torrid week for the Society of Authors in the UK, which launched a new charter for ethical AI use in the creative industries alongside international counterparts. The reaction from authors was immediate and furious: have a look at the responses on Instagram and Bluesky for a sense of it. The SoA issued a follow-up statement from its CEO a day later, which failed to take the temperature down.
For many authors, the position was as simple as arguing that there is no such thing as ethical AI usage; for this author (disclosure: an SoA member), the charter was poorly and imprecisely drafted—it confused transformer and diffusion models, for example, and seemed overwhelmingly focused on LLMs at the expense of any other aspect of the technology. That lack of clarity would make it hard to commit to. So ultimately, a compromise charter that pleased few people.
Over a hundred publishers and members of the German booksellers and publishers association used Frankfurt Book Fair to issue an open call to the German federal government European Commission and Parliament, calling on them to require AI companies to provide title-level transparency on which works had been used for training.
If you’re using Bluesky—perhaps to track the reaction to the SOA announcements—the platform’s own AI tool Attie introduced new functionality to create simple sites, polls, apps and games. Bluesky’s audience is still pretty niche, but the underlying technology, AT Protocol, is being used for some very cool things like developing a competitor to WordPress using AI tools.
On the subject of building things, the insanely clever team at Aboard used an AI coding agent to build a suite of office tools with ten applications including a word processor, spreadsheet and slide builder, where the entire file is encoded into a URL—nothing to save, just share the link and someone can access the file.
Visualising and analysing data is one of the most practical use cases for AI tools, and this week Claude launched a Dashboards feature. The difference between this and tools like Google’s Sheets Canvas is that it can integrate with third party data sources, including big, enterprise level databases such as BigQuery or Snowflake. You can click on any number to see the query that produced it. It’s also wickedly fast: I tested it by connecting it to Xero via MCP and giving a one-shot, single sentence prompt—it built a working, interactive dashboard for my business in less than five minutes. Less business focused, but it did the same for my Strava data in two minutes. In both cases, it surfaced conclusions I hadn’t picked up yet.
If you want to experiment with this, the Anthropic documentation is here, and if you’re a Claude user and you’d like a half hour online session with me to walk through this and build a prototype together with your data, get in touch.
For academics and educational publishing subscribers, OUP’s Isaac Pattis has put together a side project with an open source visualisation and analysis of over 1,600 research papers on AI and education. He also talks about the rationale for it here. It’s really interesting on two levels: for the substance of the analysis, and as an example of the kind of side project that’s possible for a single person assisted by a coding agent—even a couple of years ago, this would have been a major piece of work for a research team.
Thanks to subscriber Mathew Clayton for sharing Spotlit, a new online tool to connect literary festivals with publishers and PRs. It was vibe-coded, and has been running for six months with no technical problems. Mathew and his colleagues have two more reading focused web apps they are about to launch. In his email to me, he said, “There are lots of little problems in publishing that can easily be solved” through this sort of approach, and I’d love to highlight others people have built.
In industry news, Meta introduced new commercial terms of service for its business accounts: from 30 October, business accounts using Facebook, Instagram and WhatsApp will have to consent to Meta being able to scrape their websites for training and product improvement purposes. News publishers were the first to react to this, but given the prevalence of Meta platforms in publishing marketing and communication, this is something everyone should look at. There’s no opt-out: the only recourse if you’re dissatisfied is to stop using Meta.
OpenAI published new data on how its European business customers are using AI. The thing that stood out for me is it breaks down token use by type of model: at the end of September, ChatGPT accounted for only 45% of use, with agentic tools in the majority with 55% (ChatGPT Work was 21% and the coding agent Codex was 34%). Looking at the evolution of that data on the link above brings home how quickly agents have grown: from the start of July to the end of September, they went from a quarter of token use to over half.