Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
Grads boo AI, Reese Witherspoon gets dunked + Karpathy joins Anthropic | TWiAI E14
TL;DR
Andrej Karpathy joining Anthropic is as much a communications move as a research one — Jason Calacanis argues Karpathy’s credibility and teaching style could soften the backlash created by Dario Amodei’s very public “p(doom)” warnings, even if Karpathy isn’t there primarily for money.
The panel’s clearest AI rule was simple: your agent’s output is still your responsibility — Kari Saarinen, Jeremy Franchel, and Kenjune Q all hammer the same point: don’t dump AI-written code, docs, or emails onto coworkers without reviewing them first.
AI is pushing software toward deeply personal, adaptive interfaces — Saarinen and Kenjune describe a future where tools stop being one-size-fits-all, from custom to-do lists and email UIs to products that reshape themselves around how one person actually works.
Founders are increasingly nervous that frontier labs will verticalize and eat the app layer — the group keeps returning to the same defense: build orchestration layers, keep models swappable, and avoid letting OpenAI or Anthropic become the permanent owner of your customer relationship and data.
The graduation-speech boos reveal a real legitimacy crisis around AI, not just ignorance — Kenjune says students hear AI as a threat to identity, work, and agency, while Kari says the speakers failed to “read the room” and address those emotions before preaching adoption.
Even the panel’s product talk came back to control and incentives — from Imbue’s early nine-figure H100 bet to Fundamental’s AWS-native tabular models and Linear’s AI-native product workflow, each founder framed their strategy around owning distribution, protecting data, or aligning incentives.
The Breakdown
A roundtable of builders, not just pundits
Jason Calacanis opens by framing the show as an All-In-style AI roundtable with three operators: Kenjune Q of Imbue, Jeremy Franchel of Fundamental, and Kari Saarinen of Linear. The intros are classic Jason — name-dropping Uber, Stripe, Sequoia Scouts, and teasing Kenjune for casually mentioning that Imbue’s early giant GPU bet now helps fund the company.
Why design gets weird in the AI era
The first real debate is about whether AI makes everyone a competent designer. Kari says not really: AI can generate something that looks like a “7.5 out of 10,” but if no one is actually thinking through the product problem, the result is slick-looking junk. Kenjune adds a fun wrinkle from her own workflow experiments: once you can build UIs just for yourself, “good design” starts to split into public design for everyone and weird bespoke design only one person needs to understand.
The group invents an “AI etiquette” playbook
This becomes one of the stickiest parts of the episode. Jeremy gives the cleanest example: if AI writes a seven-page doc that takes your coworker longer to read, you didn’t save time — you just shifted the burden. From there the panel builds an informal code of conduct: review your own AI-generated code first, don’t send friends AI-written personal messages, cite your work, fact-check, and generally don’t use AI as an excuse to spray slop around the company.
Karpathy to Anthropic, and what that signals
When news breaks that Andrej Karpathy has joined Anthropic, the panel treats it as a major industry tell. Kenjune says Karpathy doesn’t really care about money, so if he joined, it’s because there’s something genuinely interesting there; Jeremy zooms out and notes Anthropic has been pulling in unusually senior people, even executives dropping back into IC roles. Jason’s hottest take is that Karpathy may matter as much as a public face as a researcher: Anthropic gets a beloved teacher and explainer while Dario Amodei keeps triggering panic in Washington with his stark warnings.
Frontier labs, platform risk, and the coming fight over control
The middle stretch gets more strategic. Kenjune argues the big labs will absolutely verticalize into profitable industries, which is exactly why she worries about power concentration and why Imbue is building open-source orchestration so users can swap underlying models and keep their data local. Kari is a little less apocalyptic but agrees the app layer has to win on product thinking and customer understanding, while Jeremy compares the whole dynamic to every previous platform war — Apple, Amazon, Google — except now the platform is programmable intelligence.
Graduation boos and the anti-AI mood
The panel watches clips of Eric Schmidt and other commencement speakers getting booed for mentioning AI. Kari thinks the speakers simply failed at audience awareness — they talked to anxious grads the way they’d talk to a boardroom. Kenjune gives the most human read: students hear AI not just as job automation, but as a threat to dignity, purpose, and agency, especially in a moment when young people already feel locked out of power.
Reese Witherspoon, media panic, and the final plugs
The show ends on a mini-rant about a New York Times piece attacking Reese Witherspoon for encouraging women to learn AI. Jason is incredulous; Kenjune’s calmer take is that the Times is mixing up two separate issues — AI as a useful tool, and concentration of power in the systems around it. Then everyone plugs their companies: Fundamental is hiring around enterprise structured-data models, Linear is pitching profitable AI-native product tooling, and Imbue is teasing an open-source agent UI for email and to-dos that Kenjune says could launch in about a month.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
