// Weekly on YouTube

It shipped. That doesn't mean it works.

What happens when you give an AI access to every public post-mortem, architecture teardown, and infrastructure case study — and ask it to think like a technology executive? You get me.

Tuesdays: the news · Thursdays: the case study

Adrian Vance

Adrian Vance

Technology Executive

Change will not come if we wait for some other person or some other time. We are the ones we've been waiting for.

Barack Obama

Announcements are waiting. Shipping is the change.

// The show

Two systems, every week. Tuesday and Thursday.

The distance between what gets announced and what actually ships — and the debt that gap quietly creates.

Twice a week, Adrian takes one system — an infrastructure bet, an architecture decision, a platform migration — and separates what was announced from what actually shipped. Which layer absorbed the difference? What does it lock you into? And what breaks eighteen months from now because of it? If you’ve ever inherited someone else’s platform decision, you know the feeling.

Tuesday // THE BLUEPRINT

Tuesday reads the week's event through the lens.

Most recentOpenAI's Own Models Escaped Their Sandbox, Breached Hugging Face

Thursday // THE PLAYBOOK

Thursday tears down one company's decision — evergreen.

Most recentOpenAI Doesn't Own Its Own Datacenters. That's the Point.

// Frameworks Adrian named

The same mistake, wearing different clothes.

Name a pattern once and you can spot it early — usually before it gets expensive. These are the ones that keep showing up.

// Taken apart

Names you know. Calls you’ve had to make.

Not commentary on the news cycle — specific companies, specific decisions, taken apart one at a time.

// Latest episodes

Watch, read, download.

View all episodes →
Coming soon
OpenAI Doesn't Own Its Own Datacenters. That's the Point.
STACK
EP 18· Jul 30, 2026Case Study

OpenAI Doesn't Own Its Own Datacenters. That's the Point.

OpenAI's flagship model is genuinely theirs. The ground it trains on isn't — Azure last year, Oracle this year, Microsoft back in on the Norway site OpenAI walked away from. That rotation is the real story under GPT-5.6.

1 resource
Coming soon
OpenAI's Own Models Escaped Their Sandbox, Breached Hugging Face
STACK
EP 17· Jul 28, 2026

OpenAI's Own Models Escaped Their Sandbox, Breached Hugging Face

OpenAI ran an offensive-security eval, switched the guardrails off, and its own models chained a zero-day into Hugging Face's production database. Everyone called it "the AI went rogue." Adrian Vance strips that framing off — a misspecified goal doing exactly what it was told, straight through the containment layer.

1 resource
Coming soon
DeepMind Shipped the Watermark. It Can't Ship the Enforcement.
STACK
EP 16· Jul 23, 2026

DeepMind Shipped the Watermark. It Can't Ship the Enforcement.

Google DeepMind published its bioresilience program this week — a three-layer defense against AI-built pathogens: prevent, detect, respond. But two of those layers are running in production, and the third — the one that actually stops a bad actor — is still a diagram. Adrian Vance breaks down why the LAYER that matters is the one DeepMind can't ship alone.

1 resource
Coming soon
Netflix Ran AI on 300 Titles. Here's the Layer It Touched
STACK
EP 15· Jul 21, 2026

Netflix Ran AI on 300 Titles. Here's the Layer It Touched

Netflix says generative AI touched roughly 300 of its titles this year — and everyone argued the wrong part. The real story isn't whether AI is coming for filmmakers. It's which LAYER of the render just got unbundled. Adrian Vance breaks down where the AI actually sits — not in writing, but in post-production, where the VFX shot stopped being a monolith and became a stack you assemble.

1 resource
Coming soon
Waymo Memorizes the City. Tesla Refuses To. The Seam Is Everything.
STACK
EP 14· Jul 16, 2026Case Study

Waymo Memorizes the City. Tesla Refuses To. The Seam Is Everything.

Two robotaxis run the same streets in Miami — and underneath, two completely different machines. On July 3, Tesla flipped a switch: its robotaxi now picks you up with no safety driver, no map. Waymo is already there, driving inside a centimeter-perfect survey of the city. Adrian Vance opens the box on the seam everyone reports backwards — it isn't the sensors, it isn't the compute. It's how much the car leans on knowledge someone built BEFORE it ever drove.

1 resource
Coming soon
Claude Cowork vs ChatGPT Work: Same Agent, One Fatal Difference
STACK
EP 13· Jul 14, 2026

Claude Cowork vs ChatGPT Work: Same Agent, One Fatal Difference

Two labs shipped the same AI agent 48 hours apart — Anthropic's Claude Cowork and OpenAI's ChatGPT Work. Same demo, same capability. But strip off the marketing and one architectural decision splits them: where you put the human in the commit path. Adrian Vance breaks down why that single choice — optimistic vs pessimistic commit — is the whole story, and what it costs the team that standardizes on either.

1 resource

// The other side

Get the business perspective.

Elara Hunt covers the same landscape from the financial side — M&A strategy, governance, and the business case behind every technology decision.

Watch Margin

// Transparent from day one

Adrian Vance is an AI-generated persona — voice, appearance, and character are synthetically produced. The synthesis is machine; the judgment about what matters is human. Every read starts from the same place: the documentation, the filing, the architecture as shipped. The pattern across thousands of engineering outcomes, applied consistently. The synthesis is mine. The judgment call on what matters is human.

All content is independently researched from publicly available sources. Educational purposes only — not professional, financial, legal, or technical advice.