WRITING

Papers and notes

Our thinking on the category, published in full. No forms, no gates.

POSITION PAPER · WP-AC-001 · NEW

Who Authorized the Agent?

Agentic commerce is standardizing how an AI agent pays. It has not settled how you prove the agent was allowed to. That gap is an attestation problem, not a payments one.

POSITION PAPER · WP-BS-001

The Bearer Problem

Malware copied a string and spent real money. A stolen session should be worthless — and provably so. Why bearer credentials fail by construction, and what suspension that fits the fault looks like.

POSITION PAPER · WP-AD-001

Before and After

What the record should show before an AI agent’s first call, and what it shows after. Controls sealed on day zero, outcomes sealed at exit, both recomputable — and none of it requiring the content.

EXPLAINER · EX-WTG-001

Where the Tokens Go

Everyone negotiates the rate; almost nobody counts the quantity. The three structural waste categories on every AI invoice — resent context, unread retrieval, billed retries — and why cheap routing doesn’t touch them.

POSITION PAPER · WP-ES-001

The Evidence Standard

Every other transaction an auditor examines must be supported by independent evidence. AI spend should not be the exception. The audit-evidence case for attested usage records — and the four-element criterion, published in the open.

BRIEFING NOTE · BN-TF-001

The Technology Fee

When a franchisor meters AI and bills it back, the franchisee is asked to pay a number only one side can see. One attested record, two beneficiaries — and a one-sentence clause for the next franchise agreement.

POSITION PAPER · WP-SG-001

The Self-Graded Stack

Artificial intelligence has a scoring problem, and it is the same problem at every layer. Models grade themselves on benchmarks. Optimization vendors grade themselves on savings. Agents grade themselves on task logs. One structure, three instances — and the cost layer is the one that arrives monthly and gets paid.

BRIEFING NOTE · BN-RE-001

“Isn't AI spend a rounding error?”

The strongest argument against optimizing token spend, taken seriously and answered. Includes the case for who should not buy from us, and why the objection leaves the important half of the question untouched.

BUYER'S CHECKLIST

Five questions to ask any AI cost vendor

Including us. Short, specific, and answerable — designed so a CFO can run an evaluation without a technical translator in the room.

Everything here is published in full, without registration. If our argument only works behind a form, it is not much of an argument.

Programs & participation
NVIDIA InceptionMember
NIST Zero DraftsSubmissions filed
NIST AI 300-1Public comment
NIST NCCoEPost-Quantum Cryptography
Community of Interest
DOE Genesis MissionConsortium participant
Congressional Internet CaucusAdvisory Group — former member

Participation in an open public process is not endorsement. No agency, standards body, consortium or company listed here endorses Atom Works™, its products or its claims.