Context Window 99
Ninety-nine weeks of doing this, more or less continuously. Forty five days to publication, and the first finished copies of my book arrived from the publisher this week. You can get a preview of it here (likes and shares appreciated). This newsletter is and will remain free, but if you’ve found it useful, please preorder the book from me, as it supports all of the research that goes into this newsletter and my public events.
Amazon announced two new AI devices this week: a wearable bracelet and a card-sized attachment for your smartphone that can record voice notes and conversations, both produced by Bee, the startup Amazon acquired last summer. The raw audio that’s captured is not retained, but is turned into notes within the Bee platform. This isn’t a new category, of course: Bee has existing competitors for both form factors. But few of them have Amazon’s marketing might, and that does increase the likelihood you’re going to experience these products, directly or indirectly when someone else brings them into your meeting.
The marketing for the devices emphasises privacy, with built-in prompts to obtain consent from people you’re talking with, and visible indicator lights to show when they’re recording. This points to the key issue, which is cultural rather than technological. I remember years ago showing some publishing colleagues the first-generation Alexa hardware, and there being some joking nervousness about having a hot mic to what was also our biggest customer. More recently, there has been a backlash against public use of Meta’s Ray-Ban glasses, with some workplaces banning them outright. We don’t allow AI notetakers on my IPG webinars because it’s so hard to manage consent from all participants. On a practical level, the real difference between the Bee hardware and what’s running on individuals’ smartphones and laptops is that the wearables are more visible. But that makes them a lightning rod for lots of personal and organisational concerns. Expect plenty of discussion and new policies in thoughtful workplaces.
If you’re responsible for that sort of policy, there’s a useful new piece of research from Careful Industries looking at issues with AI voice transcription. Although the focus of the work was the UK public sector, it makes some excellent points around tools, benefits and risks that are immediately transferable to publishing or general business settings.
The National Trust has suspended advertising on TikTok, claiming that an automated TikTok Ads feature had generated and published a series of videos that the brand had not approved. Given how many publishers use TikTok as a platform, there’s an obvious and quick compliance exercise for marketing departments: make sure that TikTok Ads Manager is configured correctly to prevent generated material being automatically published.
OpenAI’s new GPT-6 includes a feature called Intelligent UI, which builds answers to user queries incorporating interactive graphics and experiences. The examples given include cooking, travel planning and learning. For publishers in those and similar areas, this is simultaneously an instant, free, personalised competitor to instructional non-fiction and an example of the sorts of user experiences that can be developed at speed, scale and low cost. Longer term, it’s interesting to think about what a book has that the interactive version doesn’t, which means leaning into personality, aspiration and brand.
While Amazon is blocking third-party agents from its store, a consortium including Meta, Shopify, Stripe and Walmart is developing a new open standard, Personal Agent Protocol, to define how AI agents interact with third-party websites. Consortium member Walmart is leaning into this with an MCP interface and shopping profiles for customers’ AI agents, stating “we’ve always aimed to be where our customers are”—in this case, using agents like Meta’s Muse and OpenAI’s Dots.
The difference in approach between Amazon and Walmart is a tale as old as tech: companies build walled gardens or ecosystems. That has a clear implication for metadata, discovery and resourcing: do you optimise for the garden, the ecosystem, or both?
Joining Muse and Dots on the bandwagon, Anthropic has announced its own personal agent, Claude Duet, apparently built “for professionals and small business owners with more on their plates than they can track.” I feel seen.
The Society of Authors is withdrawing from the Charter of Ethical AI Use in the Creative Industries a week after introducing and championing it. As an SoA member, it gives me no pleasure to observe that this has been a textbook piece of miscommunication: an apparent absence of consultation, no pitch-rolling, a badly worded charter and a reactive response to what should have been anticipated. One author, sensing victory, responded bluntly: “publishers need to be made afraid of this level of anger too”, but I thought Nick Harkaway’s assessment was on the money: how many writers, or associated organisations, would be willing to strike over this issue. What other levers do they have?
Last week I wrote about Meta’s new commercial terms, due to take effect a week from today. Publisher groups in the United States, Europe and the UK are calling on Meta to suspend the terms, and the issue has already attracted regulatory attention. It’s going to be fascinating to see whether Meta swerves away from the oncoming headlights, or calculates it can drive straight over them.
Finally this week, Matt Locke has a superb post about AI and creative friction. He argues that some kinds of friction are straightforwardly bad, and AI can automate them away—but others need more careful thought. Good friction might slow you down and help you think through a difficult issue. It’s an excellent mix of theory and some very practical advice. I really love his conclusion on good friction (I say something similar in the book):