A CATEGORY-DEFINING REPORT
The Agent Boundary
Accountability, not capability, is the bottleneck to enterprise agent deployment. How a new layer of infrastructure turns four unanswerable questions into signed, offline-verifiable facts.
Strategic Report, Edition 1.0 · for CIOs, CISOs, CTOs, boards, regulators, and enterprise architects.
ABSTRACT
Accountability is the new bottleneck — and it is manufacturable.
Enterprises already have capable agents. What they lack is a trustworthy way to answer who acted, what data it touched, under what rule, and where it failed — their answers live in logs that are mutable, unsigned, and believable only to those who already trust them. The instant the question turns adversarial — a breach, a dispute, a regulator — that trust is exactly what is gone.
This report defines the category that closes the gap: agent accountability infrastructure — the Agent Boundary, a runtime control point at the agent's edge where identity, data control, and governance run with no change to the agent's code. These converge into one signed recording: deterministic, tamper-evident, offline-verifiable, and verifiable even with the producing engine gone.
The enterprises that learn to manufacture agent accountability — signed facts an outsider can verify with your systems switched off — will deploy agents the others cannot.
THE FOUR QUESTIONS
Strip agent governance to its irreducible core, and four questions remain.
The questions a regulator, an auditor, an insurer, and an incident commander all ask, in different words, about the same event.
WHO acted?
A signed identity bound to every action — one that survives revocation, offline.
WHAT data did it touch?
A field-level record of what crossed the boundary, bound to purpose.
UNDER WHAT rule?
Each action bound to the exact policy it ran under — effective vs. intended.
WHERE did it fail?
The first bad step, version, or input — reproducible on replay.
WHAT'S INSIDE
Eighteen chapters, across five parts.
Part I — The Problem
Why capable agents stall before production, the Capability–Accountability Scissors, and the four questions every enterprise must answer.
Part II — The Agent Boundary
Runtime identity, runtime data control, runtime governance — converging into one signed recording, deterministic and engine-absent verifiable.
Part III — The Payoff
Verify (audit, compliance, liability transfer) and Locate (root cause, MTTR). One recording, two budgets, one chain of custody.
Part IV — Economics & Architecture
The economics of accountability, competitive landscape, reference architecture, deployment models, and regulatory alignment.
Part V — The Future
Discovery, replay, control plane — and the governance layer of the agent economy.
SEE IT FOR YOURSELF
The report's central claim is re-verifiable.
Offline fail-closed revocation, engine-absent verification, first-bad-step localization — all independently re-checkable. That is the entire spirit of the category.