Writing on private AI, encrypted storage, and decentralized infrastructure.
Every AI memory product today solves "portable between models." None solve "portable away from the vendor": that means building your own exit door. Here's what that actually requires, and why nobody's shipping it.
Read article →Inference infrastructure has raised roughly $1.8B in the past two months, and "TEE-verified" is becoming the default claim on every provider's pricing page. Here's the four-link proof chain that phrase should actually mean, and the link almost everyone skips.
Read article →Alex Karp says enterprises are livid: paying for tokens, losing their IP, watching model providers capture their alpha. He's right about the problem. Here's what the infrastructure looks like when you solve it with math instead of promises.
Read article →Dario Amodei told the US Senate that open source AI is too dangerous. He's right about the risks. The solution he's proposing is regulatory capture: mandatory corporate surveillance dressed as safety. Here's the third path.
Read article →Every hosted AI API runs on legal promises. ZDrive runs on math. Here's what the difference looks like in practice — and why it matters for work you can't afford to hand to a stranger.
Read article →Anthropic's newest model retains every prompt for 30 days, no opt-out, even for zero-data-retention enterprise contracts. Here's why that backlash matters, and what ownership looks like when it's architecture instead of a policy.
Read article →PewDiePie's walkthrough of Odysseus AI hit nearly 63,000 GitHub stars. Here's what it gets right, and the structural gap between choosing privacy and actually having it.
Read article →Running LLMs on Kaggle is free — but every query flows through Google's servers. Here's what the tutorials don't mention.
Read article →