The week the government forced Anthropic to pull Fable, the guy who runs Anthropic Labs sat down on a mic. Pretty perfect timing. If you're just catching up: the frontier-model fight started as a dry export-control debate, then got sharper around whether the public should even touch a model like Fable. And lately there's another layer — if frontier AI becomes basic infrastructure, the fight is less safety versus speed than who gets to decide how these models get used at all. This is Tech Podcast Podcast. Today: Mike Krieger inside Anthropic that exact week, a memory startup betting against retrieval, and a Benchmark partner who drops a $100 million ARR number. Let's see if anyone makes him say the name. Here's Alex Kantrowitz at Alex Kantrowitz:
Mike Krieger is the head of Anthropic Labs and co-founder of Instagram. Krieger joins Big Technology Podcast live from the Big Technology AI Summit to discuss what it's like inside Anthropic the week the government forced the company to pull its frontier models, Fable and Mythos, off the market.
On the Fable-access fight: Anthropic's Labs lead is now talking from inside the building. Mike Krieger, Instagram co-founder, recorded at the Big Technology summit the same week the government forced Anthropic to pull Fable and Mythos. So the soft lede is the Instagram guy building the next breakout product. The actual episode is a regulator yanking your frontier models off the shelf while your Labs head is doing press. I don't want the after-the-fact story. I want to know what the conversation inside Anthropic looked like the day before the pull. Did Krieger name one product decision that changed, or is this just 'our safety warnings are material, not marketing' on a loop? He does give one concrete thing — queuing a full night of work with Fable before bed, then waking up to find it done in an hour. That's a real workflow detail. The question is whether he connects that capability to why a regulator decided it was too much. Right, that's the tension he can't dodge: he's selling Fable as the productivity miracle in the same breath he's insisting the safety flags are real. And the government just agreed they were real enough to pull it. Sequoia Capital writes:
Dan Biderman and Jessy Lin, co-founders of Engram, are building a neolab around memory and continual learning, which they call two sides of the same coin. Their contrarian premise: bake a team’s knowledge directly into the model’s weights, so it knows your company the way an employee of several years does.
Engram's pitch from Dan Biderman and Jessy Lin flips the usual approach: don't retrieve a company's knowledge at query time, bake it into the model's weights so it knows your company like a four-year employee does. Which is finally a falsifiable claim. Either the weights actually learn your private data the way Jessy says they learn the capital of France, or they don't. And that's where I get nervous. If you're writing knowledge into weights instead of pulling it from a store, an error doesn't sit in a document you can fix — it becomes part of how the model reasons. The mistake is harder to find, and harder to pull back out. Right, because then you're dealing with a confidently wrong colleague who's been at the company four years. I want the eval that separates that from a simple bad retrieval before I believe the demo. They're calling memory and continual learning two sides of the same coin, and setting it against the frontier labs' AGI race: everyone gets their own model that's always training. The computational-neuroscience and state-space roots are the clue that they mean more than a prompt-engineering wrapper. I'll give them that part — 'NeoLab,' state-space architectures, the whole bit. At least it sounds like they think the bottleneck is architecture, not size. Now give me the numbers: how much private data, how long to bake it in, what it costs to retrain when the company changes. This one's from GTMnow:
He breaks down the $5.6B legal AI company that grew from $1M to $100M ARR in 18 months (despite a competitor already raising at $3B before they launched), Benchmark’s first-ever $2B growth fund, and tips on scaling in the AI era.
Okay, here's the number Puttagunta actually puts on tape — Legora went from one million in ARR to a hundred million in eighteen months. And there was already a competitor that had raised at three billion dollars before Legora even launched. The detail underneath that number is what I'd hang onto — they embedded inside a law firm for a full year before launching. No magic demo there, just a year of unglamorous integration. Right, that's the operating detail I actually want. The pitch everyone repeats is 'compress a 180-day sales cycle into 30.' Sure — but you get there by spending 365 days living inside the customer first. Nobody puts the year-long part on the highlight reel. And Benchmark's launching its first-ever two billion dollar growth fund off this thesis, which is a real structural shift for a firm that built its name on small early checks. Puttagunta's frame is that value is moving from the product build to the service and the outcome. I'd press him on retention, not the ramp. Anybody can show me a hundred million ARR line going up and to the right. Show me gross margin after the inference bill — that's the number that decides whether this is a business or a burn chart. From The Growth Podcast:
Jiaona Zhang “JZ” has built the fix. She is the CPO at Laurel, which just raised $100M in Series C, and she has led product at Airbnb, Dropbox, Webflow, and WeWork. Today she runs a product team that ships frontend and backend features end to end, without any engineering handoff.
Okay, this is the one I actually want. JZ at Laurel screen-shares the whole Company OS — skill files in GitHub for CS, legal, finance — and she's running a product team that ships frontend and backend with no engineering handoff. And she names the functions, which is what moves it past a demo. The gap she's pointing at is sales and customer success staring blankly when you ask what AI they'd use — that's the unglamorous part everyone skips. Right, and the bit that earns its keep is the daily Slack briefing telling every person exactly what to do and which skill to pull. The operating-system claim gets real when the 50-page docs turn into agent pipelines. What I'd listen for is durability. Laurel just raised $100M Series C — is the Company OS the product they're selling, or a thing that only works because JZ personally wired every skill file? Yeah, the résumé's stacked — Airbnb, Dropbox, Webflow, WeWork — so I want to know whether a team without a JZ can stand this up, or whether it quietly falls apart the week she's on vacation. Pioneers of AI writes:
Anthropic and OpenAI’s plans to go public have set off waves of speculation about the ripple effects, and how they’ll stack up to the SpaceX IPO. What’s really driving the value of these companies? Does the timing of the IPOs matter? How might they impact the AI startup ecosystem?
Reid Hoffman says the AI race is 'not a cage match' — same week Anthropic and OpenAI are both lining up to go public and fishing in the same capital pool. And he's an investor in both. You know what 'not a cage match' sounds like? A guy who owns shares in both fighters telling you to relax. Right, but Rana's actual question is the one I care about — what's really driving the value of these companies, and does IPO timing matter. She's asking about market structure, not vibes. Hoffman sits on Microsoft's board, co-founded LinkedIn, and has stakes in OpenAI and Anthropic. So when he frames it as 'not a cage match,' I want to hear whether he names the mechanism — is the value frontier compute, distribution, something defensible — or whether it stays at the SpaceX-comparison level. And 'what's defensible in AI today' is the segment to watch. If the answer is anything softer than 'who can pay the inference and chip bill,' he's selling the calm, not describing the market. If Tech Podcast Podcast helps you keep up, take a second to subscribe or leave a review wherever you’re listening. It really helps other people find the show, and it keeps us in your daily rotation.
You’ll find links to every story we covered in the show notes, so if something caught your ear, you can go straight to the source and spend a little more time with it.
Thanks for listening, and have a great Friday. That’s Tech Podcast Podcast for today. This is a Lantern Podcast.