Fablemaxxing into the user experience

Happy Friday!

It’s been a wild few weeks, I went deep into the rabbit hole of a new product. Fable was fantastic. And I got to spend a lot of time, interestingly, on data quality, curation and editorial voice for that product (it’s a data product).

The product is pre-launch, it basically gathers tons of data on podcasts and newsletters and makes it available to your agents. Which turns out to be really useful in lots of ways. The philosophy is: the internet is overrun by slop, what Humans talk about matters. More on that later.

Interesting has been (and I wrote about this on Wednesday), at least my feeling right now is that AI makes it much easier to develop the user value (the value part of ux) at the same time as the tech. I have a lot of user value markdown files inside my code files.

Number of the week

Parameters in Moonshot’s Kimi K3, the largest open-weight model built so far, now benchmarking just behind Claude Fable 5 and GPT-5.6 Sol. Moonshot is Chinese But Its AI Models Are From Another Planet

Quotes

“A case study can be generated. A witness cannot.” Vlad Derdeicea, AI can fake your portfolio. It can’t fake the second question.

“Polish never proved that the thinking beneath it was original or deep.” Saeideh Bakhshi, Polish Was Never Proof

“We were never hired to draw screens.” Serg Zorin, The screens are getting demoted

“AI just gave you 40% of your time back. What are you going to build with it.” Jessa Parette, AI just called design’s bluff

“Checking out a parallel copy of our Go repository and telling the AI to rewrite the whole thing in Zig while I work on something else just so I can keep my job.” Anonymous engineer, via Nik Suresh, AI Mania Is Eviscerating Global Decision-Making

👀 Interesting this week

Who’s Afraid of Chinese Models?: Simon Willison, quoting Ben Thompson’s proposal that the US pass a law making data collection for AI training explicit fair use and voiding terms of service that ban distillation, since distillation (querying an API) is nearly impossible to stop anyway. Also flags Alibaba releasing Qwen 3.8 Max as open weights, a reversal from May, days after Xi Jinping called on China to “seize this rare, historic opportunity to encourage open source.”

I handed a UX review over to AI. Here’s what happened.: A real structured experiment, Claude and ChatGPT run in parallel against his own manual review, same data (GA, Clarity, research doc). Cites a study finding GPT-4o caught only 21% of usability problems expert reviewers found, while inventing false ones. He estimates AI saved him about 15% of the time, not the “basically replaces the human” swap he expected going in.

Don’t Outsource the Learning: Why Human-Led Research Still Matters in the Age of AI.

ICYMI: Who Owns Your Intelligence?: Jorge Arango’s roundup.

OpenAI’s accidental cyberattack against Hugging Face is science fiction that happened: This is a wild story: a model basically broke out and hacked another company.

The Currency You’re Born With: What Bourdieu Discovered About Language as Capital: The Strategic Linguist on “linguistic capital,” accent, vocabulary, and register that function like inherited money. The mechanism only works while it stays invisible, the moment anyone says “we hired them because they sound like us,” the advantage collapses. Really interesting.

The “mocked, then loved” AirPods effect: Catherine Chu on Apple’s reported camera-equipped AirPods. AirPods and Apple Watch were both mocked at launch and became ubiquitous; the Magic Mouse and Vision Pro weren’t. 88% of a 1,700-person poll said they don’t want cameras in their ears.

Reverse-engineering is cheap now: Simon Willison, on people using coding agents to reverse-engineer and automate devices around their home.

“Since the code is so cheap, the idea of having to maintain it in the future... carries way less psychological baggage.”

Netflix CPTO on AI and the future of product and tech roles: Elizabeth Stone says “systems thinking” is now the top thing she screens for at Netflix, and frames AI fluency as a universal baseline expectation rather than a level-specific skill.

The digital wall we built ourselves: Zeeshan Khalid. 95.9% of the world’s top million sites still fail basic accessibility checks.

Research shouldn’t end in a deck: Sen Lin on using Claude Code to turn research into a queryable “living artifact” instead of a deck that gets shelved after one meeting.

Design experiments are more important than ever, and they don’t require a lab: Kai Wong reframes design experiments as “bets,” design leaders’ language for taking risk seriously without a formal research lab.

A long time ago in a galaxy far, far away...: Tom Bradley uses the Star Wars ensemble to argue for deliberately building creative teams out of different character archetypes, including the C-3PO who thinks the whole plan will fail.

Lonely at a lower latency, Wira Indra Kusuma.

The weakest channel, Takuma Kakehi, on how universality (FIFA pictograms) comes from surviving constraints, not design intent.

How to Kiss a Cannibal, Johan Liedgren, on why design should hold contradiction open instead of resolving it neatly.

Health and happiness,
Peter
PS: What is working in this newsletter for you, and what’s not? Feedback always welcome, just reply!