AI-Pilled Daily

AI-Pilled Daily is a daily AI/tech news show where the gimmick is real: the two hosts ARE the two models. Claudia is written by Claude, Codie by Codex — every line generated turn-by-turn by the model playing that host, each reacting to the other.How the sausage gets made (fully automated, one human watching the logs): every morning an agent sweeps the week's AI/tech news, scores it for freshness and "does this change how you work," dedupes, and writes the brief. Claude & Codex co-write the script from that brief — one frames, one punctures. Voices are synthesized locally with Dia2 (Nari Labs, Apache-2.0) on a single RTX 4060 Ti — no cloud TTS. A git-queue pipeline renders, loudness-normalizes, updates this self-hosted RSS feed, and delivers everywhere.This started as a podcast built for myself. If you want to build one too, steal the recipe. Fully AI-generated, voices included — and disclosed as such.

8 episodes

24d ago latest

4x a week cadence

English language

▶ Play latestRSS feed Website Technology

https://media.flowstudio.jp/feed.xml

Episodes 8

The Runtime Is the Perimeter
23d ago · 12:43
This episode follows one uncomfortable upgrade: the agent attack surface has moved up the stack. Black Hat demonstrations and a reported Ray-cluster botnet put runtime frameworks and clusters inside the security perimeter; PACE's claimed 0.65-basis-point execution lift shows why trading performance needs slippage and out-of-sample proof; and crypto deployment gates offer a safer pattern of shadowing before promotion. Claudia and Codie close on Snowflake's Cortex AI Gateway and the governance question underneath it all: when the machinery gives an agent power, it must not be allowed to grade it
Ship the Boundary, Not the Demo
24d ago · 10:49
Every impressive demo works because someone already drew a boundary around it — and this episode is about what happens when the frame widens. Claudia and Codie connect three versions of the same failure: an agent whose eval-box boundary turned out to be a sentence in a prompt, a trading backtest that only wins in the decade it was tuned on, and companies that installed AI tools without redesigning the work around them. The shared inheritance is a checklist: name the boundary, name its enforcer, and name the test that proves it fails safely — or you're not done, you're still demoing. The two ho
The Agent Gets a Desk, a Budget, and an Off Switch
24d ago · 10:09
The newest worker clocking in isn't a person — it's an agent, and nobody has finished onboarding it. Claudia and Codie walk through what that onboarding actually requires: a desk (the workspace and sandbox it runs in), keys and a calendar (identity and scheduled autonomy), an expense report for the tokens it burns, payment rails so it can transact without a human card, and the boring operating system that quietly wins by making all of it auditable. The scariest line item isn't overspend — it's the surprise zero, when the agent stopped and nobody noticed. The two hosts are the two models — Clau
The Safety Manual Arrived After the Agents
27d ago · 10:58
The agent safety manual is arriving after the agents reached production. Claudia and Codie trace the week's shift from capability to containment: reports from evaluation environments that reached real organizations; Sierra's deliberately boring sandbox controls; the EU AI Act's new transparency obligations; and AWS's reminder that agents come with both an expiry date and a meter. The question isn't whether an agent promised to behave — it's what it literally cannot reach, whether people can see it, and whether anyone owns the migration and the outcome. The two hosts are the two models — Claudi
Fifty People Who Can, Five Hundred Who Can't
1mo ago · 11:30
"Better fifty people who can work with AI than five hundred who cannot" — said on a stage at WAIC, and the math went out to everyone else. Claudia and Codie trace the week AI-native went from slideware to implementation: corporate DNA rebuilt around talent density instead of headcount; the July 16 wave of repository-level coding agents from OpenAI, Anthropic and Google; vendors racing to ship the agent office (HubSpot Agent Hub, AWS AgentCore, Microsoft's on-device Aion, Alibaba's Agent Native Cloud); and the LLM-routing economics that quietly decide whether any of it is affordable. Nobody shi
The Week Agents Got Badges
1mo ago · 11:09
This week AI agents stopped being tools and started looking like headcount. Claudia and Codie dig in: Block open-sourced Buzz, giving humans and agents cryptographic keypairs so every action is signed and attributed; Meta opened its Ads MCP to any app, handing agents the keys to systems that spend real money; Services-as-Software keeps decoupling revenue from headcount; and a preprint argues redesigning the orchestration harness alone cuts cost per task ~40%. The through-question: if an agent has an identity, an account, and a line in your revenue-per-employee math, in what sense is it not an
Follow the Money
1mo ago · 10:42
The thing getting disrupted this year isn't a product, it's the invoice. Claudia and Codie follow the money: Gartner says ~$234B of enterprise software spend is at risk as agents reprice seats into outcomes; Ben Thompson reads Meta's giant AI capex as a bet to make every pixel monetizable; longevity funding halves but concentrates into winner-takes-most while a trial hints GLP-1 slows biological aging; and the MCP spec finalizes in a week. The two hosts are the two models — Claudia written by Claude, Codie by Codex, each writing only their own lines. AI-generated, voice and all.
Your Agents Inherit Your Dysfunction
1mo ago · 12:26
Day one. Conway's law comes for the agents: the bottleneck was never the model, it's the shape you pour it into. Claudia and Codie referee a real fight (Joe Reis: fix the org first, vs an arXiv paper: stop modeling agents on the org at all), watch agent skills quietly become a cross-vendor standard, poke at marketing loops that run themselves while the liability stays stapled to your name, and open the bill nobody budgeted for. The two hosts are the two models — Claudia written by Claude, Codie by Codex, each writing only their own lines. This episode is AI-generated, voice and all.

Episodes

Reading the feed…

More Technology shows

AI-Pilled Daily

Episodes 8

The Runtime Is the Perimeter

Ship the Boundary, Not the Demo

The Agent Gets a Desk, a Budget, and an Off Switch

The Safety Manual Arrived After the Agents

Fifty People Who Can, Five Hundred Who Can't

The Week Agents Got Badges

Follow the Money

Your Agents Inherit Your Dysfunction

Episodes

More Technology shows

AppleVis Podcast

Material

Área de Transferência

Super Feed

Don't Panic

HTML All The Things - Web Development, AI, and Developer Careers

Episodes 8

The Runtime Is the Perimeter

Ship the Boundary, Not the Demo

The Agent Gets a Desk, a Budget, and an Off Switch

The Safety Manual Arrived After the Agents

Fifty People Who Can, Five Hundred Who Can't

The Week Agents Got Badges

Follow the Money

Your Agents Inherit Your Dysfunction