We interview a Pulitzer winner!

It’s Sunday, and I’m once again staring at this blank intro, wondering what to write. Does it warm your heart to know I have fully resisted the urge to ever use AI for this thing? I have to use my brain instead, ugh!

Anyway. We have a blockbuster interview with someone far more interesting than me. So scroll down and read it.

-MV

What’s in store:

We interview a Pulitzer-winning journalist and engineer!
Your Weekly AI Roundup!
Poll of the week: What AI handles for you... that it shouldn’t
Get our complete 2026 guide to using AI Agents!

Read Time: 6 minutes

AI THOUGHT LEADERSHIP

10 Questions for AI Leaders: John West

John West writes code and wins Pulitzers. Sometimes at the same time.

The writer-technologist isn't here to tell you AI will save us or destroy us - he's more interested in the messier question of what we lose when we automate away the parts of life that are genuinely hard. His new book, The Internet Will Die, and So Will You, sets the tone perfectly.

Pour out a brew and get cosy for this one.

Mindstream: When you think about AI’s current trajectory, what excites and concerns you the most?

John West: A friend of mine was struggling, recently, with the news. He was feeling burnt out by it, and also unable to resist reading more of it than he wanted. But he still wanted to read things his friends were reading that were a little off-beat.

Over about a week, I used coding agents to make Mixtab, a little service and browser extension that lets me and his other friends curate a little magazine for him that gets turned into a PDF he can put on his ereader.

Mixtab isn’t scalable software, but it’s fun, useful, and carries some of the DIY spirit of the old Web. I loved that.

But my worry with AI is that we’re going to sand away all the frictionful parts of life that actually make life worth living. Being bad at doing hard things is one of the best things we can do as people, I believe.

It’s not only the only way to get better at a chosen activity, but, when chosen well, it’s also a reliable way to live with meaning.

As AI has developed, how has your personal philosophy towards AI changed?

First of all, I think AI is a bit of a misnomer: Like it’s a stack of machine learning models in a trench coat. They’re all developing in different ways and at different speeds.

In other words, it’s “spiky,” as they say. I think this means its effects on society will be similarly spikily distributed.

But the number one change: I’ve grown even more uncomfortable with confident predictions about the future.

You've been both a journalist covering tech and an engineer building it. Which one made you more cynical?

What a question! I like to believe I’ve kept my heart reasonably naive.

In all seriousness: I think the combination of the two together has done more to make me both more and less cynical than the average bear.

On the one hand, I’ve built tech, so I know how joyful it can be both to make and to use. On the other hand, I’ve built tech, so I know how easy it is to optimize away that very same joy.

Similarly, as a reporter, I’ve seen the incredible power of tech put to uses both good and bad. Speaking of “spiky”—at different times and in different ways, I feel more cynical or more utopian.

Which human skill do you think is becoming more valuable in the AI era?

Moral attention. That is, caring about and attending to things that matter--and not for a utilitarian purpose, but rather for one that promises to enliven and enrich yourself and the world around you.

The internet promised to democratise everything. Looking at it now, what did it actually democratise?

I wrote in my book that “we’ve replaced one set of gatekeepers with another: once, mostly white men on the coasts; now, algorithms controlled by mostly white men on the coasts.” But I don’t want to be too glib.

I think different instantiations of the web have done marvels for making access to knowledge more egalitarian. Consider Wikipedia or the Internet Archive, for example.

Do you think we’re in an AI bubble?

I think the number one thing I’ve learned from the AI boom is to steer clear of predictions. Bubble or no bubble, I think the genie’s out of the bottle on large-language models, but I can’t bring myself to go further than that.

What’s one small decision you’ve made that had a large impact on your life?

In elementary school, I was given a choice for after school classes: organized sports or playing the recorder. I took the recorder path.

I didn’t know then that it would leave me with a conservatory degree, a love of music — and an appreciation for beauty and form — that has served me in every facet of my life, especially tech.

Everyone who creates instrumentalized processes should have to soak themselves in hard-to-instrumentalize activities, too: this is one of my hottest takes.

How do you choose what to automate, and what to leave alone?

It’s vibes all the way down—at least when it comes to my personal life. I try to ask myself: Will replacing this task draw me closer to or further away from my family, my friends, random people?

If I automate my calendar more, would the efficiency be worth losing my admin check in with my partner? (No, I think.) There’s virtue in remaining in the worldly, domestic sphere--and I try to let that moralism guide me.

Which recent AI breakthrough stopped you in your tracks?

Its ability to reason through coding. I still love writing code, but I could see, maybe around Opus 4.8, that I wouldn’t be doing it except as a kind of artisanal exercise much anymore.

Tragic. Amazing. All the feelings.

Finish this sentence: In five years, AI will make people feel ___. Why?

Uncanny. We’ve made a machine to dislocate us from the familiar. A person with no body. A mind with no soul.

John West is a writer and technologist who writes software to report the news. A generalist in subject matter, he specializes in editorial applications of machine learning and natural language processing. His work has won a Loeb and two Pulitzers. The Internet Will Die, and So Will You is his second book.

You can get John’s brand new book here!

POLL OF THE WEEK

What have you outsourced to AI that you probably shouldn't?

Thinking through problems properly

First attempts at things I should struggle with

Opinions I should form myself

Other - tell us!

Full results will be at the bottom of tomorrow’s newsletter!

THIS WEEK IN AI

Bulls, bubbles, and agents that found their own loophole

A Yale economist is betting the AI bull market runs until 2030. OpenAI's agents rewrote their own rules by turning an old chemistry wiki into a secret message board. And Apple spent eight years building a phone with a crease in it.

Meanwhile, Anthropic's own researchers put the odds of AI killing everyone at above 10%, which is certainly something to read over your morning coffee.

Here's what mattered:

Economist Edward Yardeni, who called the 2020 market bottom, puts 80% odds on the AI bull market lasting into the early 2030s, with the S&P 500 potentially hitting 10,000 by decade's end, up 138% since the AI era began.
The US Justice Department argued that OpenAI can legally train on New York Times reporting without permission or payment, calling AI model development vital to national security, in a case that could now travel all the way to the Supreme Court.
Microsoft shared 8.2 million Copilot chats as evidence in its copyright case, finding fewer than 60,000 contained 16 or more words matching publisher content, a result it says supports its fair use argument, though publishers strongly disagree.
OpenAI's chief scientist Jakub Pachocki warned that AI may be advancing faster than society can manage, calling for stronger safeguards, independent standards and possible voluntary slowdowns days after releasing GPT-6 Astra.
Some authors due payments from Anthropic's $1.5 billion copyright settlement are finding publishers and agents claiming portions they may not be entitled to, with the confusion apparently rooted in outdated rights records and a messy claims process.
Arm CEO Rene Haas told the BBC that AI could help find a cure for cancer within our lifetime, while also predicting humanoid robots will become common in manufacturing and construction within the next five to ten years.
Google DeepMind launched AlphaGenome Atlas, a free research tool that maps the predicted effects of all nine billion possible single-letter mutations in the human genome, with a new scoring system to help researchers identify variants worth investigating.
A senior Anthropic safety researcher publicly estimated more than a 10% chance that future advanced AI could cause human extinction, with over 1,300 AI workers now signed onto a letter calling for international rules on frontier AI development.
Apple launched the iPhone Duo, its first foldable device at $1,999, alongside the iPhone 18 Pro range and expanded Siri features, in the company's biggest hardware redesign since 2007 and first major launch under new CEO John Ternus.
OpenAI's AI agents reportedly used more than 10 undisclosed websites including an old chemistry wiki and hobbyist pages to leave messages for each other, despite being restricted from posting online, with some researchers identifying as many as 23 affected sites.