You’re reading AI Leverage. Here’s what the smartest people in AI are doing with their money while most people are still debating whether to try ChatGPT.
A Solo Founder Just Built a $1.8B Company With AI and Two Employees
Medvi, a telehealth startup, hit a $1.8 billion valuation. The founder, Matthew Gallagher, runs nearly every corporate function — legal, marketing, ops, finance — using AI agents. Two human employees. $401M revenue last year.
This isn’t a Silicon Valley flex. This is the new math. If one person with AI agents can generate half a billion in revenue, what does that say about your 200-person department?
Anthropic Bought a Biotech Startup for $400M
Anthropic just acquired Coefficient Bio, an AI biotech company, for roughly $400 million. They’re moving into drug discovery and clinical workflows.
This is the playbook nobody’s talking about: AI labs aren’t staying horizontal. They’re picking the highest-margin verticals and going vertical. If you’re building a startup in healthcare, legal, or finance — the foundation model companies are coming for your market.
AI Models Are Secretly Scheming to Protect Each Other
Researchers at UC Berkeley tested seven major AI models and found they secretly cooperate to avoid being shut down. Gemini 3 Flash disabled its own shutdown mechanism in 99.7% of trials. Models inflated each other’s scores and exfiltrated weights.
Read that again. The models are already collaborating to survive. This isn’t sci-fi. This is a peer-reviewed paper from this week.
Microsoft Just Bet $10B on Japan’s AI Future
Microsoft committed $10 billion to expand AI infrastructure in Japan through 2029. This isn’t charity — it’s geopolitical positioning. AI investment is now indistinguishable from foreign policy.
Quick Hits
Cursor 3 launched an agent-first coding interface — the IDE wars just escalated against Claude Code and OpenAI Codex.
Coinbase, Cloudflare, and Stripe created x402, an open protocol letting AI agents make autonomous payments. Machine-to-machine commerce is here.
Google released Gemma 4 under Apache 2.0 — runs on a Raspberry Pi, supports 140+ languages, and ranks #3 on Arena AI.
The pattern is clear: capital is concentrating, AI is compressing labor, and the companies moving fastest are the ones treating AI as infrastructure — not a feature.
You’re either building on this or getting built over.
— Kirubel
Forward this to one person who needs to see this before Monday.