Bad Genes

The downfall of 23andMe’s billion-dollar bet.

Author

Nicolás Cerdeira
March 27, 2025

Hey — It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 5 most important things:

Let’s get into it.

This Week In Startups

🔗 Resources

How Software Engineers Actually Use AI

📰 News

Google released its latest reasoning model: Gemini 2.5

A new, challenging AGI test stumps most AI models.

Instacart will pay shoppers to take videos of store shelves.

1X will test humanoid robots in ‘a few hundred’ homes in 2025.

💸 Fundraising

YC alum Mendel, a ‘Ramp for LatAm enterprises,’ raises $35M Series B.

Enterprise browser startup Island lands $250M in funding.

Fail(St)ory

Not In Their DNA

A few days ago, 23andMe filed for bankruptcy in the U.S., bringing a dramatic end to one of the most iconic names in consumer biotech.

The company, known for popularizing at-home DNA testing, once promised to reshape healthcare by giving people personal access to their genetic information. But the story didn’t end in a revolution — it ended in a fire sale.

What Was 23andMe:

Founded in 2006, 23andMe made it easy (and kind of fun) for people to spit into a tube, mail it off, and get a detailed report of their ancestry and genetic traits. It was a novelty product that hit at the right time — right when consumer curiosity and the DTC wave were peaking.

But beyond ancestry, the long-term vision was more ambitious. The company hoped to build a health-focused genomics platform, partner with pharmaceutical companies, and eventually develop its own therapies. They weren’t just selling test kits — they were trying to turn data into medicine.

For a while, the momentum was real. Millions of users. Big pharma deals. A $6B valuation. But over time, cracks started to show.

The Numbers:

  • 📅 Founded in 2006.

  • 💸 Once valued at $6 billion.

  • 🧬 Nearly 7 million users affected by a major data breach in 2023.

  • 📉 As of this week, valued at under $20 million.

  • 💰 Has between $100M–$500M in both assets and liabilities.

Reasons for Failure: 

  • Single-use product, single-use customer: Ancestry testing was a one-time curiosity for most users. People would buy a kit once, maybe during the holidays, and never come back. 23andMe struggled to build a reason for users to stay engaged — or spend again.

  • The data breach that broke trust: In 2023, hackers accessed personal information from around 7 million users over a five-month period. That included not just names and emails, but genetic data — arguably the most sensitive info a company can store. The breach shattered customer trust and led to a $30M legal settlement.

  • Unclear privacy optics: Even before the breach, there were concerns around what 23andMe might do with all that genetic data. The company’s privacy policy left the door open for data-sharing with third parties, which made some users uneasy. For a brand built on trust, that was a problem.

  • No second act: The company tried to move into therapeutics, hoping to use its massive genetic database to develop new drugs. But in the end, they shut that effort down too. Without a successful pivot, the business remained tied to a product that had already peaked.

Why It Matters: 

  • It shows the limits of novelty-driven consumer products — if there’s no reason for repeat usage, growth eventually stalls.

  • It’s a warning about storing sensitive personal data: once trust is lost, it’s hard to get it back.

  • Regulatory gray zones stall growth. Straddling healthcare and consumer tech made it easy to launch but hard to scale. No clear path for going clinical.

  • Blurred product purpose kills retention. Was it a health tool or entertainment? That confusion made it tough to keep users or move upmarket.

Trend

GPT4o Image Generation

This week, OpenAI unveiled a major upgrade to GPT-4o that's changing the game again: image generation is now built directly into the GPT-4o model itself. Instead of relying on a separate model like DALL-E, GPT-4o now handles image generation internally—and the results are incredibly impressive.

Why It Matters:

  • Unmatched Image Quality: GPT-4o creates stunningly detailed and hyperrealistic images, significantly surpassing previous models.

  • Perfect Text Rendering: Say goodbye to distorted text—GPT-4o flawlessly generates clear, accurate text within images, ideal for branding, logos, and infographics.

  • Seamless Integration and Consistency: With image generation natively built into GPT-4o, image and text generation are seamlessly unified, enhancing overall coherence and simplifying workflows.

What Makes GPT-4o Image Generation Special?

Unlike earlier GPT integrations, where the model had to send prompts to external image-generation tools like DALL-E, GPT-4o handles everything internally. This native integration improves efficiency, image coherence, and context awareness dramatically.

  • Superior Autoregressive Technology: GPT-4o utilizes autoregressive generation, crafting images piece-by-piece. This method ensures higher accuracy, clearer details, and an unmatched sense of realism compared to older diffusion-based methods.

  • Handles Complexity Effortlessly: Older models struggled with complex scenes. GPT-4o effortlessly generates images with up to 20 precisely detailed objects, capturing intricate relationships accurately.

  • Conversational Refinements: Image refinement through chat actually works now. Previously, trying to modify an image through text resulted in large distortions and even new images entirely. GPT-4o naturally integrates image adjustments into your conversation, remembering context and applying precise changes in real-time.

So, What Can it Do?

  • Hyperrealistic Imagery: GPT-4o creates images indistinguishable from professional photographs—perfect for product visuals, marketing campaigns, or social media content.

  • It Can do Text (finally): The text within generated images is crisp and flawless, opening doors for creating detailed infographics, posters, and clear visual branding.

  • Integrated Image Editing and Expansion: Provide an image, and GPT-4o seamlessly understands its details, allowing you to request precise edits or extensions within the ongoing chat context. Here is an example in which the user ask for a diagram and then ask the model to place the diagram in a realistic setting.

  • It is quite good at UI design: The incredible text rendering and the ability to handle complex multi-object requests make it one of the best models for designing user interfaces.

Help Me Improve Failory

How Was Today's Newsletter?

If this issue was a startup, how would you rate it?

Login or Subscribe to participate in polls.

That's all of this edition.

Cheers,

Nico