One lab's safety branding allegedly helped put seven-point-four billion dollars in a Chinese rival's pocket. Somebody's gotta show their work on that one. If you're just joining, the frontier-governance fight has moved out of abstract safety warnings and into real access decisions around powerful models. Anthropic pulled Fable and Mythos off the market at government direction, and Anthropic Labs lead Mike Krieger has been explaining the product shift — and why the company treats its safety warnings as material. This is Tech Podcast Podcast. Today: the DeepSeek-Anthropic causal claim, a billion-dollar CTO who went back to writing code, and whether European models can actually guard your data. Let's see who's got receipts and who's just got vibes. The Information writes:
DigitalOcean CEO Paddy Srinivasan talks with TITV Host Akash Pasricha about token optimization, caching, and how developers are managing the CPU boom. We also talk with Tech & Politics Reporter Leo Schwartz about Senator Mark Warner's upcoming discussion draft targeting AI agent guardrails, and Asia Bureau Chief Jing Yang about DeepSeek's massive $7.4 billion capital injection triggered by Anthropic’s secret Mythos model preview.
Okay, the headline says Anthropic's Mythos preview "forced" DeepSeek's $7.4 billion raise. That's a hell of a causal verb for a model basically nobody's seen. Right, because until now Mythos was mostly a policy story — why a regulator decided some capability was too much. Jing Yang is adding the competitive fallout: one unreleased preview from Anthropic moves billions on the other side of the planet. And it gives us a cleaner number. We were working off a fuzzy fifty-billion figure on DeepSeek before — now it's a concrete $7.4 billion, with a trigger attached. I want someone to pull apart the mechanism here. Did Anthropic's safety-forward posture scare capital toward a rival, or did the preview just reset everyone's sense of where the frontier is? Those are very different stories. And that same rundown has Senator Warner drafting guardrails for AI agents specifically — not AI broadly, the autonomous ones. First time someone in Congress aimed at the thing that acts instead of the thing that just talks. From How I AI:
Eddie Kim is the co-founder and CTO of the payroll and HR platform Gusto, which just crossed $1 billion in revenue and serves more than 500,000 small businesses. Recently he did something most CTOs don’t: he went back to writing code. With three other engineers and one designer, Eddie built Gusto Cofounder, a net-new AI product, from zero code to a tier-one launch in 10 weeks.
Okay, this is the one I actually wanted all week. Eddie Kim, Gusto's CTO: the company crosses a billion in revenue, serves half a million small businesses — and he goes back to writing code. Three engineers, one designer, zero to a tier-one launch in ten weeks. And it's a senior leader doing it, not a junior dev being handed Claude Code. That's why it feels different — the CTO of a billion-dollar payroll platform decides the four-person team is the unit of work again. The detail I love: the trash-can method. Write a full PR, review it, then delete it — and let that be the product decision. That's the opposite of every planning-doc PM culture I've ever seen at a company that size. I want to hear the eval-first workflow. He's using Claude Code to fix real customer bugs with evals up front. At Gusto's scale, on payroll, every bug fix can touch somebody's paycheck. Right, and here's my standing gripe — a designer with no engineering background hit the 94th percentile for shipping code. Great line. Shipping... what, though? Reviewed by whom? On a payroll product, I'd like the answer before the applause. From PMF Show:
Render CEO Anurag Goel — employee #8 at Stripe — built a $1.5B cloud platform by obsessing over developer experience, word-of-mouth growth, and founder-market fit. This episode is essential for early-stage founders navigating the gap between 'people like it' and true product market fit. Key takeaway: real PMF shows up as organic word-of-mouth, not just satisfied users.
Four years to ten million ARR, no marketing spend, no freemium tier. In dev infra, that's basically heresy — everyone else is buying logos and burning runway to fake the curve. And the launch team was four people: one designer, three engineers, founder included. Six million developers later, the support team is still under ten. That's the stat that does it for me. Hundreds of thousands signing up every month from word-of-mouth, and they're keeping support in single digits because everyone there can actually read a stack trace. Goel was employee number eight at Stripe — so he watched developer experience compound into a business once already, then went and rebuilt the playbook. Instead of model differentiation, the pitch is a product good enough that people drag their coworkers onto it. Yeah, the slow burn is the proof here. He's saying 'people like it' and real PMF are different things, and the tell is whether they bring friends without you paying them to. DevSecOps Talks writes:
If episode 103 asked whether you still need AWS, this one asks a sharper question: why would you hand your company's secrets to a US-owned model at all? Mattias, Andrey, and Paulina continue the European-sovereignty conversation with Mark Shine (Co-Founder & CTO of Stockholm Kubernetes platform Ankra) and Pawel Piwosz (Developer Advocate) and Filipe Berti (Agile Team Coach) of Finnish cloud provider UpCloud — this time on EU LLMs and whether they're actually good enough yet.
Okay, finally, a sovereignty take with a cost attached. Mark Shine's pitch is basically this: an EU data center run by EU people can still hand your IP to the US government under the CLOUD Act, and you'd never even know. That's the right scary point — geography isn't sovereignty if a US-owned operator can be compelled. It's a real second-order cost nobody prices in when they pat themselves on the back for 'EU-hosted.' But then he overreaches. He invokes a US court decision to claim anything you upload to a model becomes public domain — and the case, U.S. v. Heppner, turned on legal privilege, not trade-secret loss. He's smuggling a scary headline past a ruling that doesn't say it. And that matters, because the number that actually drives this episode is the half-a-billion-a-month model bill that makes self-hosting pencil out. If you're going to argue 'own the car instead of taking the Uber,' make the case on the cost math, not a misread court opinion. Right — the Uber-versus-car framing is actually useful. At a half-billion a month, you stop renting capability and start asking what the EU tooling gap costs you. That's the honest question. The public-domain thing is chum. If Tech Podcast Podcast is part of your daily routine, subscribe wherever you're listening so you don't miss an episode. And if you have a moment, leave a review — it really does help other people find the show.
What we're watching next: Senator Mark Warner's discussion draft on AI-agent guardrails. We'll be looking for what actually makes it into the proposal.
As always, we've put links to every story from today's briefing in the show notes, so if one caught your ear, you can dig in there. That's Tech Podcast Podcast for today. This is a Lantern Podcast.