Notes from
the workbench.
Long-form writing on zero-knowledge cryptography, rollup architecture, account abstraction, and the infrastructure that keeps the open web open.
Your Sealed-Bid Auction Is Not Sealed.
A sealed-bid auction on the EVM is not actually sealed: you fake it with commit/reveal, a deposit, and a forfeiture rule. I built the same auction in Solidity (239 lines) and in Daml on Canton (561 lines) to show what code disappears when the ledger keeps the secret itself, and where the cost moves instead.
ERC-4337 is Not Enough.
ERC-4337 proved account abstraction could ship without a hard fork. Three years into production, the cost of that choice is visible: a permanent gas premium, a bundler oligopoly, a continuous audit surface on the EntryPoint contract, and a delegation cousin in EIP-7702 that already has its own phishing category. Native AA at the protocol level is the only architecture left that can close those gaps.
State Channels, Round Two.
ERC-7824 shipped as a draft in January 2025 and Yellow Network is building a real crypto clearing business on the reference implementation. The protocol layer is ready. The opinionated layer above the SDK is where the next year of state-channel apps gets shaped.
Andrej Karpathy Just Killed My RAG Pipeline.
On April 3, 2026, Karpathy posted a gist with no new framework, no benchmarks, and no infra. It got 16 million views on X and 5,000 stars. The reason is that it quietly showed the RAG stack most of us were paying for was never needed at our scale.
ERC-8004: The Standard That Decides Whether AI Agents on Chain Is Real.
Draft EIP created August 13, 2025. Audited contracts live on 20+ networks January 29, 2026. About 20,000 agents on the Identity Registry. The spec is good. The next twelve months decide whether that matters.
Statelessness: The Final Frontier of Decentralization.
A full node needs 700 GB today and grows 30 GB a year. A 2026 validator wants 4 TB of NVMe. The fight over Verkle vs binary-plus-STARK is the fight over whether a phone can run a node in 2029.
WTF is OP_CAT? Why Should We Care?
A twelve-line opcode Satoshi silently removed in 2010, a Schnorr signature quirk Andrew Poelstra noticed a decade later, and the reason every serious Bitcoin layer-2 team is quietly lobbying for a soft fork.
The Case for Sovereign Rollups on Bitcoin.
Chainway raised $14M on a paper titled 'A Sovereign ZK Rollup on Bitcoin.' The rollup they are shipping is not sovereign. That pivot is the argument of this piece.