Wow - 100,000+ people engaged with the launch of Mitosis Labs yesterday. Today, we're launching our first product: "Cortex."
Over the past 7 months, PJ and I helped more than 6,000 people get their first AI agents. A lot of them were business owners. Every single one of them had the same dream: to accelerate the growth of their business and run parts of their company with AI. Even we wanted it.
So why are businesses spending millions on AI and seeing no return on investment?
The truth is, you can't just throw decades of your company's data at AI and expect it to work.
Let's take the example of Brandon, one of our customers, who runs a marketing agency. His business has been accumulating data for a decade. Good data and bad/legacy data, mixed together across email, documents, CRMs, spreadsheets, and chat history. When Brandon handed all of it to his AI, it couldn't tell the difference. So it answered confidently and wrongly. That's a hallucination, and inside a business, hallucinations are catastrophic because they can lead to sending a client a report with the wrong numbers, or quoting a price from a rate sheet you retired two years ago.
So the missing ROI on AI spend is not because the models aren't smart. It's because they're reasoning over polluted data.
Today, cleaning that data is a boring, billion-dollar consulting industry. Human labor, lots of compute, and years of work.
Mitosis Cortex does it automatically. You connect the sources your business already uses. Cortex works out what's true, disregards what isn't, and traces every answer back to its raw source, so you can always see exactly why your AI said what it said. This helps your existing AI tools actually become usable inside your business. When your AI stops guessing, you stop double-checking it.
More importantly, it cuts your inference costs, because your data only needs to be processed once. From then on, it stays up to date as new data comes in. That creates a compounding effect: your company's brain, built on clean data, ready for AI.
Mitosis Cortex has scored state of the art on LongMemEval, the standard benchmark for AI memory. And we published our full testing harness, so you don't have to take my word for it:
▎ Cortex: 91.7%
▎ Supermemory: 85.2%
▎ Zep: 75.1%
▎ Mem0: 73.8%
Now, this next part is IMPORTANT, because by the time businesses realize it, it might be too late. So I want to bring it to your attention.
Alex Karp (CEO of Palantir) said it on CNBC: when you plug your company's data into a frontier lab's model, you're not buying intelligence. You're paying them to hand over your IP, your personal data, and your edge.
Ask Figma, whose stock dropped 52% when Anthropic released a competing product.
Cortex runs on private inference. Open-source models that never train on your data, on-prem if you're regulated. Every data point carries its own privacy policy. Additionally, your data is stored in a secure vault, encrypted end to end, and quantum-hardened in transit. You decide, deterministically, what AI can see and what it can never share.
I spent the last 10 years in cybersecurity getting ready to build this. In 2016, in Chicago, we started a think tank with an Ethereum co-founder, launched a network at a $100M valuation, and we were building coding agents before coding agents were cool. Cortex is that decade compressed into one product.
We run our entire company on Cortex. In fact, our entire CRM is hiding somewhere in the video below. Go find it!
If you're a business owner who is spending a lot of money on AI, or who wants to get ahead of the game and avoid the same mistakes, please reach out. We're happy to help you.
2 weeks later, our CRM of openclaw and hermes agents is now over 50,000 members, with full contact information for nearly all of them
the agent community is exploding
@theo This is the biggest untapped opportunity in the ai space. The fact that Cloudflare still hasn’t launched their cli tool says something.
I just want to be able to prototype products without ever touching a dashboard. Is that too much to ask?!!
Thanks for building this.
We just ran 100 agents all weekend
5 different model providers, 20+ different models used
Each agent has the option to spawn a new specialized agent or break tasks down
Results:
>20k tasks completed
1000+ files created
2000+ PRs
Everything reviewed by adversarial debate 🚀