Context Window 73
It’s been a big couple of weeks for AI image generation and design tools: with major updates from Canva, Google and Anthropic, users are spoiled for choice. The new tools speak to a trend I’m increasingly seeing—and write about below: the baseline for AI use is shifting from competent prompting to confidently delegating to agents. The exam question for publishers is getting used to managing systems of agents rather than using individual tools.
Google’s AI small business marketing tool Pomelli is now available to customers in the UK and Europe. It reads your website, produces a brand kit grounded in your existing content, and then suggests social media campaigns and product images. I’ve been playing with it for a couple of days and it’s an impressive tool for smaller publishers without creative agency support.
It’s important to stress that Pomelli is an experimental, Google Labs project, a category of product that hasn’t always been known for longevity. For a more fully-featured approach, Canva just announced a major upgrade to its AI functionality, including iterative, prompt-based design and layered object intelligence—effectively being able to take just one part of a flat image and edit it.
Continuing the design and branding theme, Anthropic released Claude Design, a new tool to generate app and website layouts, slide decks and infographics. Designs can be exported, edited in other tools like Canva, or sent to Claude Code for integration into live projects. It’s available on Claude Pro tier and above.
However, on that subject Anthropic also experimented with removing Claude Code from its Pro pricing tier this week, leading to a backlash from users. The reality is that it is probably underpriced for most users: even moving to the more expensive Max tier would still cost less than half a day of a developer’s time for considerably more output. But a permanent change would hit light or occasional users. My advice to price-sensitive publishers: do your experimentation now.
Away from general design tools, Google also introduced two new AI agents designed for academic publishing: PaperVizAgent is a development of an experimental project for drawing academic figures formerly known as PaperBanana, and ScholarPeer is a peer reviewer agent that evaluates academic papers. Relevant publishers might find them worth trying, though note the not so prominent disclaimer that these are experimental projects and should not be relied on for publishing decisions.
As I was writing this newsletter on Thursday, OpenAI released the highly anticipated GPT-5.5: after months of Anthropic setting the pace for model development, the new model promises increased efficiency as well as greater capability. I haven’t had a chance to test it yet, but early reviews are positive.
Social reading platform Readwise released a new skill for OpenClaw, allowing the AI agent to access bookmarked and highlighted content. This is on top of an existing MCP integration with Claude and ChatGPT. Being able to query highlights and notes from reading platforms like xigxag and Kobo is a neat feature. Thanks to subscriber Cameron Drew for drawing this to my attention.
There’s been a lot of debate about editors using AI models to appraise manuscripts, with the Author’s Guild putting out a statement that prior consent should be required before loading submissions into consumer-facing chatbots.
The differentiation between general tools and professional services like Storywise with proper guardrails is important: I commented that the irony is that publishers that use dedicated tools with proper communications and disclosure will come in for criticism from many authors by drawing attention to it, but they will certainly have better security than a publisher where individuals use personal AI accounts in an unstandardised and unsanctioned way.
I’ve had a busy week with a lunch and a lunch-and-learn trends 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, and I wrote a piece about the changing patterns that I am seeing: the baseline shifting from competent prompting of LLMs to confidently delegating tasks to agents, generational patterns in AI use, and the impact of AI on small and solo firms.
In thinking through how best to use agents, I really liked this piece from technologist Betty Junod on the role of communications and context: “When you ask an agent to operate in your environment, it needs what your team never fully codified: the context of why a decision was made, the shared language of your domain, the goal behind the request—not just the request itself. [What] we never did well for our human colleagues is missing in what we feed the machine.”
I need to pack for Chicago. I have editorial comments back on my book, and a long flight ahead to go through them. Have a good weekend.
This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.