Blog

Notes from the build.

Short essays on AI evals, agent products, and the cryptography under them. Every one ends in something you can check: a repo, a demo, or a number.

01 Essays

The cluster is the result. xAI is the exception.

In a 22-model T=0.1 political-questionnaire cohort, all 20 non-xAI models across nine US- and China-headquartered providers answered from the same equality–liberty region. Release drift was family-level, temperature had no common direction, and the country comparison depended on one provider.

An image is not a sentence

From an analytical-philosophy obsession to VSON: a notation where every claim about an image declares every argument it takes, viewpoint included.

Trick the agent all you want. The vault does not care.

Assume prompt injection will happen. Maestro puts ten on-chain checks between a hijacked agent and the money, and is honest about what an attacker still gets.

Three agents, one signature, none holds the key

How Chorus runs an AI-agent committee on FROST threshold signatures and ERC-7710 delegation: no key to steal, no authority that was not explicitly given.

App islands: an edge you can rent is not an edge

AI collapsed the cost of building your own internal tools. Notes on skipping the CRM in San Francisco, and why the next moat is the internal codebase.

Why we ship products from cryptography papers

There is a long gap between a zero-knowledge paper and something a person can click. Notes on how we close it.

Hiding zero-knowledge latency is a UX problem

Proofs take seconds. Players will not wait seconds. What designing Mina Mastermind taught me about masking compute.

AI agents need wallets. They also need a kill switch.

Off-chain guardrails are bypassable middleware. The case for putting agent spending policy on-chain.

02 Thoughts

@yamancan →