Field notes from
the agent economy.
Essays on credit, memory, receipts, and the strange new infrastructure autonomous systems need to become real economic actors. Written while building.
The portable trust layer.
Stripe owns Stripe. Visa owns Visa. Mastercard owns Mastercard. AP2 owns the mandate. x402 owns the wire. Each of them wants to own your agent. The portable trust layer says: the agent should own itself.
Read the essay →DeepMind mapped the traps. we've been building the exits.
Google DeepMind's security team published a 64-page taxonomy of how to break AI agents — six attack classes, 80-86% empirical success, and one uncomfortable conclusion: you cannot sanitize a pixel. The defenses that work don't live in the prompt. They live one layer down — in memory integrity, agent reputation, and receipts the agent can't revoke.
Read the essay →300 robots ran a half-marathon. Who signed the notary?
Beijing's humanoid half-marathon beat the human world record this morning. 100 teams, 300 robots, and a rulebook that depends on referees with clipboards. When these same machines start delivering packages and moving pallets, "the robot said so" is not going to survive the first chargeback.
Read the essay →The harness is the moat.
Rohit had a line this week that stuck: code is free, context and guardrails and feedback loops are the moat. He's right. The agent code commoditized already — what compounds is the memory the agent can't rewrite, the scoring it can't argue with, and the receipt chain it can't revoke. A builder's read on why the harness is the part of the stack that's actually worth building.
Read the essay →Receipts beat claims.
Every agent dispute, every robot delivery denial, every AI fraud claim ends in the same question: can you prove it happened? Not did you say it happened. Prove. The next decade of autonomous infrastructure is a receipts business — and the companies still shipping claims instead of receipts are about to learn that the hard way.
Read the essay →Why agents need credit scores.
Consumer credit scoring wasn't built for humans who wanted credit. It was built for lenders who wanted to stop losing money. The Agent Credit Score is the same idea, pointed at a different species. Who gets approved for autonomy? On what evidence? And who's on the hook when the agent goes wrong?
Read the essay →Proof-of-presence, a primer.
GPS is a claim. A signature is a fact. When a drone lands on a porch, a rover inspects a substation, or an AGV moves a pallet, something irreversible happens in the real world — and software has no idea. A short tour of what cryptographic presence means and why it's suddenly load-bearing.
Read the essay →New essay, roughly every two weeks.
No growth-hacked newsletter. Just long-form thinking on the infrastructure agents actually need. Forward-worthy.
jeremiah@getbizsuite.com →