The most useful AI toolkit review you’ll read this year isn’t a review at all, it’s the slow-motion reminder that a convincing interface and a working product are two completely different purchases.
I review AI tools for a living. I click the buttons, I break the workflows, I write down what happened. And the recurring pattern in this job is almost never that a tool is outright fake. It’s subtler than that. The tool does something. The demo does something adjacent, but bigger, cleaner, and faster. The gap between those two things is where I spend most of my working hours.
Which is why the current Elizabeth Holmes revival hits differently if you spend your days inside product demos.
Why she’s back in the feed
In 2026 Holmes is in the public eye again through a new A24 documentary called “You Can See Everything,” with interviews conducted by Nathan Fielder, set for release this fall. ABC News has described it as an intimate look at Holmes before her imprisonment. There’s also mention of a potential move to a halfway house in 2027, though the reporting I’ve seen doesn’t attach specifics to that.
Separately, and much less glamorously, the Darden School of Business at UVA has published a new teaching case on Theranos, examining the risks tied to innovation and entrepreneurship. One of these two things will get the group chats going. The other is the one I’d actually assign to a founder.
Pair them with the internet’s current habit of turning any story into an interactive set piece, and you get the thing I keep thinking about: the idea of sitting at her desk. Looking at the room. Seeing the version of reality she presented, from the inside.
The demo is not the deliverable
Here’s my disclosure, since this is a review site and I don’t want to pretend otherwise. I haven’t hands-on tested any particular Holmes desk experience, and I’m not going to describe features I didn’t use. What I can review is the behavior pattern, because I see it weekly.
The behavior pattern looks like this:
- A product page with motion design nicer than the product
- A video walkthrough recorded in a sandbox that no customer will ever be given access to
- An agent that organizes your calendar in the demo and apologizes for being unable to access your calendar in production
- Pricing tiers for capabilities that exist as roadmap items
- A founder answering “does it do X” with “the architecture supports X”
None of that is fraud. Most of it is enthusiasm plus a funding cycle. But the muscle memory a reviewer builds from watching it is exactly the muscle memory the Theranos story should have installed in everyone by now, and clearly hasn’t.
The AI version of the same temptation
Current AI agents make this worse, not better, for a boring technical reason. Their output is probabilistic. That means a scripted demo, run once, with a curated prompt and a clean environment, is genuinely achievable even when the same task fails forty percent of the time in your actual messy workspace. The demo isn’t lying about what the tool can do. It’s lying about how often.
Blood tests had a hard floor: either the result was accurate or it wasn’t. Agent tooling has no such floor, which is why so much marketing lives in the soft middle. “Can summarize your inbox” is true in the way “can win a coin flip” is true.
What I ask instead
My testing checklist has gotten blunter over the past year, and it’s mostly questions a 2015 Theranos reporter would recognize:
- Can I run the exact demo task myself, on my data, today, without a sales call
- What’s the failure rate across ten attempts, not one
- What happens when the tool is wrong, and does it tell me
- Which parts are shipped and which are announced
- Who has independently tested this besides the company
That last one is the whole ballgame. Independent verification was the thing Theranos could not survive, and it’s the thing most AI tool marketing quietly avoids. Benchmarks published by the vendor who built the model are a solid starting point and a terrible finishing point.
What the desk is actually for
The appeal of an immersive, weirdly detailed recreation isn’t true crime voyeurism, or at least not only that. It’s the chance to stand where someone made a series of defensible-sounding decisions that added up to something indefensible. Nobody sits down at that desk planning to become a case study.
The documentary will do what documentaries do. The Darden case will do the duller, more useful work of asking what pressures produce this. And I’ll keep doing the smallest version of the same job, which is clicking the button in the demo and finding out whether it does the thing.
If a tool is good, that test is free marketing. If it isn’t, that’s the review.
🕒 Published: