What Caught My EyeCulture · Brands · Design · BuildingWeekly Links by Mike LitmanIssue #007Ten Links. No Filler.Agents with Taste & Google's 75% AI CodeCurated, Not AggregatedWhat Caught My EyeCulture · Brands · Design · BuildingWeekly Links by Mike LitmanIssue #007Ten Links. No Filler.Agents with Taste & Google's 75% AI CodeCurated, Not Aggregated
Weekly Links
What Caught My Eye
Curated links on culture, brands, design and building.
Week of 24 April 2026 · Issue #007
↓ ten links. no filler. your weekend reading, sorted.
AI & BuildingCulture & BrandsDesign & ProductStrategy & BusinessWild Card
Emil Kowalski's essay on how to transfer taste into an AI agent. Not "prompt engineering" in the LinkedIn sense; a proper working framework for encoding aesthetic judgement into the system so the agent's output feels like yours, not like everyone else's.
This is the most useful piece I have read this year on what AI-native design actually looks like in practice. Taste was always the moat. The new craft is translating it into a system prompt, a reference set and a feedback loop, so an agent can carry it. Read it, then go audit what your tools are defaulting to.
read twice, bookmark, steal the framework
Send to:the design lead who thinks AI tools are producing generic output because "the models are still too dumb"
Sundar Pichai confirmed on the Q1 earnings call that roughly 75% of the new code Google now ships is generated by AI, up from "more than a quarter" just 18 months ago. Engineers have shifted from typing code to reviewing, directing and chaining agents.
The number itself is almost beside the point. What matters is what happens to the org chart once 75% of the craft is agentic: fewer but more senior engineers, teams measured on review throughput, and the middle of the pyramid steadily disappearing. Every CTO should be modelling this before their CFO does.
the reorg is coming
Send to:the VP of Engineering who still thinks "AI assistance" means faster typing
Meta has confirmed it will monitor employee clicks and keystrokes across internal tooling to generate training data for its AI systems. Framed as a way to capture tacit workflow knowledge at scale, it is also, unambiguously, surveillance at the desk.
This is the trade-off every large employer is about to face in public. If AI's next moat is proprietary process data, the easiest place to farm it is your own workforce. The companies that figure out how to do this with consent, compensation and clear boundaries will win the talent war. The ones that do not will make the headlines.
keystrokes as training data
Send to:your head of people ops, about five minutes before the employee questions hit
A LessWrong post running through 10 genuinely creative personal uses of AI from the past month. No "summarise my meetings." Actual texture: negotiating with a landlord, triaging a medical question, pressure-testing a decision. The kind of stuff most operator-level AI posts skip.
Most AI-use-case lists are written for audiences. This one reads like a working notebook, and that is exactly why it is valuable. If your team's "AI adoption" still looks like meeting summaries and email drafts, you are leaving most of the value on the table. Pick three of these and trial them this week.
stealable, usable, now
Send to:the colleague who says "I have tried AI, it's fine" and then closes the tab
The Economist on how Waterstones has become one of the British high street's genuine success stories: trained staff who actually read, curated local ranges, inviting stores, and a deliberate refusal to out-Amazon Amazon. The case for traditional retail principles, executed properly.
Every retailer now talks about experience. Waterstones is one of the few actually delivering it, and the model is almost embarrassingly old-fashioned: hire people with taste, trust them to curate, make the store a place worth being in. The lesson is not "books are special." The lesson is that specificity beats scale when your customer has infinite options elsewhere.
old-school beats algorithm
Send to:any brand leader whose retail strategy is still "what if we copied DTC"
Former PM Rishi Sunak, now advising on AI policy, has gone on the record: AI is already reducing job opportunities for young people, and entry-level roles are the most exposed. The graduate-to-senior ladder that underwrites most knowledge work is the first rung AI is pulling out.
You can argue about the timing, but the direction is not in doubt. If the entry-level roles disappear, the question nobody has answered is where future senior operators come from. Firms that invest early in building juniors through AI-native apprenticeships, rather than replacing them, will own the talent in five years. This is a strategy brief, not a doom headline.
rebuild the ladder
Send to:any CEO sponsoring a "graduate programme" they have not reviewed since 2023
New Epoch AI analysis of Ipsos data shows a sharp income split in AI adoption: Claude users skew strongly to higher-income households, Meta AI users skew the other way, and ChatGPT sits somewhere in the middle. The assistants are sorting consumers faster than anyone is admitting.
This is the first real data on what "AI as infrastructure" looks like across class lines, and it should reshape every go-to-market conversation. If your ICP is high-earning knowledge workers, they are in Claude. If it is mass-market consumers, they are in Meta AI. The tool someone uses is now a demographic signal as clean as the phone in their pocket was ten years ago.
the AI you use is the new postcode
Send to:the growth lead building an AI product and still treating "AI users" as one audience
Stratechery's reading of Tim Cook's tenure as Apple CEO, built around the argument that his actual skill is timing, not vision. When to ship, when to wait, when to buy, when to kill. A defence of operator-as-strategist in an era that worships founders.
The Silicon Valley narrative has spent a decade undervaluing Cook because he is not Jobs. Thompson's piece is the best rebuttal I have read, and the broader point applies well beyond Apple. Timing is a strategy discipline in its own right. The operators who know when to move often beat the visionaries who know what to build.
timing is the strategy
Send to:anyone who thinks their company needs a "visionary" when what it actually needs is a better operator
Chicago Booth economist Alex Imas flips the usual question. Forget what AI replaces; ask what becomes scarce when cognition is cheap. His answer: attention, agency, trust, context, and people who can tell the difference between a good answer and a plausible one.
This is the cleanest reframing of the AI-and-jobs conversation I have read. Scarcity, not automation, is the right lens. Every career, product and brand decision for the next five years comes down to: am I investing in things that get rarer as AI gets better? Print this one out. Actually.
the scarcity list is the roadmap
Send to:the strategist who still frames AI questions as "what gets automated"
Plans have been approved to turn Leicester Square into a Piccadilly-style advertising spectacle, giant digital boards and all. London now has two "the lights" locations instead of one, and the implications for media buyers, brands and the city's night-time image are bigger than they look.
Out-of-home is supposed to be in structural decline. Somehow the most expensive digital boards in the country keep multiplying. The real story is about scarcity again: truly iconic, unskippable ad inventory is one of the few formats AI cannot deflate. If you can stand in front of it, you can charge for it. Worth the click, or just enjoy the idea.
two Piccadillies now, apparently
Send to:the media planner who keeps being asked why OOH budgets are going up, not down
telegraph.co.uk
Spot something I should feature?
Reply to any issue or tag me on LinkedIn. If I use it, you get the credit.
your name in lights (well, in Caveat font)
don't miss next week's picks
Get notified when a new issue drops
A short email every Friday with the link. The reading happens here.
No spam. Unsubscribe anytime. Powered by Buttondown.