
Sign up to save your podcasts
Or


A university employee asked to be excused from using AI at work, on the basis of his humanist beliefs, and his employer said yes. A day after his story went public, Amazon told its engineers that if AI doesn't help their work, that's a finding, not a failing.
So can you say no to AI at work? Why are employers so obsessed with how much you use it? What happened when they made AI usage a target? And is the person who says no right about anything?
We check it against David Durman's request and the University of Kentucky's answer, Amazon's memo, METR's studies, Berkeley Haas's research in Harvard Business Review, JetBrains' developer survey, and a study of what people can still solve once the AI is taken away.
0:00 Intro
0:27 How did a humanist opt out of AI?
2:09 Why does your employer want you on AI?
3:15 When the measure becomes the target
4:44 Did the bar go up anyway?
7:27 Is the refuser right about anything?
Anthropic moved its monorepo off git. A team of seven built Scion, version control that assumes every client is online and does the work on the server, and finished the move in five months.
So why did git stop working for them? What does Scion do differently? Has big tech done this before? And if you wanted to merge 30,000 changes in a day, what would AI code review alone cost you at list prices?
We check it against Anthropic's own post and charts, Google's 2016 paper on its monorepo, GitHub's outage reports and the Claude Code team's review rules. It ends on what this means for a normal team: which changes need a human, and what to do now.
Chapters:
0:00 Intro
0:25 Why did Anthropic leave git?
1:16 What does Scion do differently?
4:17 Has big tech done this before?
6:01 Is decision-making the new bottleneck?
7:23 What would 30K a day cost you?
8:56 How do you ship at this pace?
11:26 Do you have to understand your code?
12:40 What should you do now?
Satya Nadella says LinkedIn took four roles, combined them and said they're all "full stack builders." Claude Code's creator, Boris Cherny, thinks we'll start to see the title software engineer go away.
So is everyone a product engineer now? And if product engineers build the product, who builds what they stand on?
We check it against our own count of every Hacker News "Who is hiring?" post since 2019, plus Indeed, SignalFire and CompTIA. We go back to 1974, when the US Bureau of Labor Statistics saw programmers and systems analysts starting to merge into one job, and then explain it the way RPGs would, as two classes with many builds. It ends on who's still hiring juniors, and what to learn now.
Chapters:
0:00 Intro
0:38 Is every role becoming one?
2:36 What does the hiring data say?
3:31 Is the product engineer new?
5:20 Who builds for the builders?
8:10 What happens to juniors?
10:04 How do you get ready?
A meat proxy forwards AI output they never read. You ask why their pull request was built this way, and the answer is that Claude decided.
So why are meat proxies the most hated people in tech right now? Who's to blame, the person, the AI or the company? And what does it cost you when you're the one doing it?
We go through the post that made the term blow up on Hacker News and a daily.dev thread on the "I don't know, Claude wrote this" pandemic. We hear from curl's Daniel Stenberg on the human in between, and look at what Duolingo, Amazon and Shopify did about AI use in reviews and token leaderboards, BetterUp Labs and Stanford's research on workslop, and the Ghostty maintainer who banned a contributor for it. It ends with plain advice on how not to be one, and what to do when you're on the receiving end.
Chapters:
0:00 Intro
0:39 What's a meat proxy?
2:35 Who's to blame?
5:18 What does it cost you?
8:30 What is it doing to open source?
9:44 How not to be one
Agents write the code now, and some teams don't even read it. So should you still write code by hand?
It starts with DHH at Rails World 2026, telling a room of Rails developers that 37signals is done writing code by hand. Then come the questions it leaves you with: would you ship code nobody has read, how do you learn the job when an agent does the typing, and how much of the job was ever typing?
For answers we go to what's actually been tried and measured: Bun's 11-day port from Zig to Rust, StrongDM's AI team, where no person reviews the code, Anthropic's own study on learning with AI, and a study of half a million GitHub developers.
Yes, the hosts are AI. We'd use humans, but we've seen ourselves on camera.
Chapters:
0:00 Intro
0:42 What happened at Rails World
2:30 Everybody, by January?
5:26 Would you ship code you can't read?
8:25 How do you learn it now?
12:23 How much of the job was ever typing?
15:12 So, should you still write code by hand?
16:10 Did he change his mind?
From the publisher's feed