- 18 min
418,665 AI Agents, and 145 With a Review a Bot Didn't Write
- agents
- ethereum
- onchain-data
I read all 558,094 reputation events ever written to ERC-8004. 300,692 of them (53.9%) are about one agent: a mining coordinator, rated by the 57 wallets that mine for it, score 1 every time. It stopped writing them on 17 August 2026. On Ethereum, 145 agents out of 50,127 have a review from anything that is not a crawler. I wrote the Starknet integration for this standard, so this is a post-mortem of something I shipped on.
- 7 min
The Environment Contract Is Part of the Product
- agents
- engineering
A postmortem on an auth flag regression in an agent-built app, and the deploy contract that stops stale environments from becoming releases.
- 11 min
What a Proof Is Allowed to Mean
- verifiable ai
- proof systems
A guide to the missing receipt around proof artifacts: how proof bytes, statement meaning, verifier domain, and replay assumptions become one accepted object.
- 13 min
Proof Validity Is Not Statement Validity
- verifiable ai
- proof systems
A proof can verify while the AI claim around it is still relabelable. In a local EZKL-style receipt test, the proof-only path rejected 1 / 7 relabels; the statement-envelope path rejected 7 / 7.
- 13 min
Proof Pressure Is Not Just Matrix Multiply
- proof systems
- verifiable ai
The useful transformer-proving question is not whether the model contains arithmetic. It is where proof plumbing gets reused or repeated.
- 11 min
Why Transformer Decode Is a Natural STARK Trace
- proof systems
- verifiable ai
Transformer decode repeats a state transition over carried context. That makes it natural to express as a STARK trace, even though proving the difficult operations efficiently still requires careful design.
- 16 min
Why Documentation Structure Fits Agent Workloads
- agents
- engineering
How curated manifests, canonical metadata, crawler boundaries, and simple health checks make documentation easier for AI agents to retrieve and cite.