Knowledge Center/ Research
Provy Research · The publication library

The definitive body of work on operating autonomous AI

A numbered series of evergreen, citable publications. Each one defines a concept, argues why it matters, and stays at the level of principle rather than product. Read as a whole they are the reference an enterprise team returns to when a green dashboard is not enough to trust an agent. Start with the canonical piece, then follow the path in order.

Featured research
No. 001 · The canonical publication · 13 min read

Decision Assurance

The discipline the whole library is built on. It names the problem in a line and points to the full treatment, then lays out the commitments that make an autonomous decision trustworthy: measure the real outcome, define a good one in advance, hold the evidence, and keep verifying. It sets Decision Assurance apart from monitoring, evaluation, guardrails, and governance.

Read the canonical publication →
Field guides

The plain-language walk-throughs. Start with the overview, then follow one run all the way through.

Field guide · 6 min read

How Provy Works

Trace in, proof out. What goes into Provy, what it does with it, and what comes out, with one diagram. Read this first for the picture before the theory.

Read the field guide →
Field guide · 5 min read

How Provy Checks the Work

Follow one run start to finish: the guess Provy makes up front, the real result that comes back, and how lining the two up becomes a trust score. Plain language, one worked example.

Read the field guide →
The publication series

Read the path in order

Publications 001 through 007 are the learning path. Each builds on the one before it, from naming the problem to a framework for locating your own organization. This is the same order the Start Here path follows.

No. 001 · 13 min read

Decision Assurance

The discipline itself: the commitments that close the gap, the vocabulary that ties them together, and how it differs from monitoring, evaluation, guardrails, and governance.

Read →
No. 002 · 12 min read

The Enterprise Confidence Gap

The distance between what an AI system appears to do and what it actually achieves, why that gap is widening, and what closing it demands.

Read →
No. 003 · 13 min read

Why AI Observability Isn't Enough

The tools you already run watch the machine run. Correctness lives in the result. The category error, laid out tool by tool.

Read →
No. 004 · 13 min read

Outcome Intelligence

Measuring each decision against the real result it was meant to produce, and the difference between an estimated outcome and a reconciled one.

Read →
No. 005 · 12 min read

Runtime Evidence

Confidence you cannot show a third party is not enterprise confidence. Why the record that ties a decision to its outcome must be held at the moment, not reconstructed later.

Read →
No. 006 · 12 min read

Continuous Verification

Trust is a property you keep, not a gate you pass once. How to keep checking that a system still does the right thing as the world moves under it.

Read →
No. 007 · 7 min read

The Enterprise Confidence Model

The five stages an organization moves through toward trusting autonomous AI, and how to find the one you are in and the next move to make. The end of the path.

Read →
Foundational reading

The conceptual grounding

The two publications that ground everything else: how autonomous systems actually fail, and the gap that failure opens between activity and outcome.

No. 008 · Foundational · 13 min read

AI Failure Modes

A taxonomy of how autonomous decisions actually go wrong, beyond hallucination, and which of the six modes a green dashboard will never show you.

Read →
No. 002 · Foundational · 12 min read

The Enterprise Confidence Gap

The problem the whole discipline answers: why enterprises can build autonomous agents faster than they can trust them, and what the distance between the two is made of.

Read →
Operational research

The frameworks that put it to work

Move from principle to practice: a model for locating your organization, the life of a single decision, and the working rules for running the discipline day to day.

No. 007 · Framework · 7 min read

The Enterprise Confidence Model

The five stages from hope to assurance, and how to find the one you are in and the next move to make.

Read →
No. 010 · Framework · 7 min read

The Decision Lifecycle

The life of one autonomous decision across six stages, and where confidence is won or lost at each one.

Read →
No. 011 · Framework · 7 min read

Enterprise AI Operating Principles

The working rules for running autonomous AI with assurance, for the platform team that has to put the discipline into practice.

Read →
Most read

Start with these

If you read only three, read these. They carry the core of the argument, from the discipline itself to why current tooling falls short and what to measure instead.

No. 001 · 13 min read

Decision Assurance

The discipline that closes the gap between what an agent did and what it achieved.

Read →
No. 003 · 13 min read

Why AI Observability Isn't Enough

Why a green dashboard can sit on top of a real failure, tool by tool.

Read →
No. 004 · 13 min read

Outcome Intelligence

Measuring results instead of activity, and the gap between estimated and reconciled.

Read →
Recently updated

All publications refreshed, July 2026

Every publication in the library is now at version 2.1. The July 2026 pass added a publication number and named author, an executive summary, in-practice scenarios, and a signature diagram to each paper, on top of the abstract, takeaways, references, footnotes, and citation each already carried. See the change history at the foot of any paper for the detail.

Read the discipline, then see it on your own agents

The publications describe what confidence in autonomous AI takes. Provy is how an enterprise puts it into practice: every agent decision measured against the real outcome, with the evidence to prove it.

Get a demo

New to the topic? The Start Here learning path walks the same numbered series in order, with the concepts and insights that support each step.