Enterprise agent architecture
The Contextual Agentic Enterprise brings people, models, agents, and systems together around shared business context, explicit authority, and evidence for consequential action.
Enterprises possess enormous amounts of data, knowledge, process history and technical capability, yet struggle to assemble the right context, evidence and authority at the moment a consequential decision must be made.
You have to decide what the agent is allowed to touch, and defend that decision to people who will not accept "the model is usually right" as a control.
You own what runs in production. The interesting failures are not hallucinations; they are actions taken with authority nobody meant to grant.
You have to review a system whose behaviour varies by design. That is possible, but only if the evidence was designed in rather than reconstructed afterwards.
Available nowPaperback and kindle
From systems of record to governed AI agents and enterprise outcomes
Most enterprise AI stalls after the demonstration. This book is about what comes next: designing the context, authority, controls and evidence that let an agent take part in real work.
Watch — ninety seconds
The Next Enterprise Era. Why a working model was never the hard part, the four rings that contain an unreliable one, and what the controls cost.
There is no speech in the film, so there is nothing to caption. Everything it says is written on screen, and this is that text.
Your model works. That was never the hard part. Then the real questions arrive: which version of the policy did it use; could it reach another customer's data; who authorised that tool call; what happens after a timeout; can a document tell the model to ignore its controls; how do we reproduce a decision made three months ago. Not one of them is a model question.
An assistant produces information that a person then acts on. An agent acts. Consequence, not output — that is the whole change.
So the enterprise encloses it. Probabilistic reasoning sits at the centre, free to be wrong. Around it, deterministic control holds invariants, state and commit authority. Around that, governance holds identity, tenancy, purpose and policy. Outermost, an evidence and outcome loop records what happened and whether it worked. Four rings, each answering a question the model cannot.
The enterprise does not make the model reliable. It makes the model's unreliability survivable.
Three proposals cross one authority and control boundary. A £40,000 credit adjustment is committed, inside policy and limit, with evidence written. A contract variation with an unseen clause is deferred, because no rule covers it and a named human decides. An export including another tenant is refused, because it breaks a tenancy invariant, which is never negotiable. Same model, three outcomes: the boundary decided, not the model.
And the control layer costs about one 358,000th of a single forward pass. Control is not the expensive part. Not having it is.
A connected reading path from enterprise architecture to everyday business practice. Start with the architecture, explore security and software delivery, or choose the beginner-friendly small-business companion for practical, almost no-code workflows.
Book 2 of 4
The security
To be released
Planned for the end of October 2026. Dates can move.
Book 3 of 4
The delivery
To be released
Planned for the end of October 2026. Dates can move.
Book 4 of 4
The small-business companion
To be released
In development; release date to be announced. Dates can move.
The enterprise does not make the model reliable. It makes the model's unreliability survivable: probabilistic reasoning sits inside deterministic control, governance and an evidence loop, and authority lives only outside the model.
Models that return a label from a fixed set, with probabilities, are easier to route, test and afford than free text. They are still model judgments. Calibration, abstention and routing are your design work, and the authority check sits outside the model.
A single model call, a fixed workflow and an agent loop are three different designs with three different cost profiles. Choose by whether the path can be known in advance, and keep authority outside all three.
Traces show that each step succeeded. They don't show whether the steps formed an acceptable process. Mining agent runs against invariants does.
A team working on how organisations give AI agents authority over real work without losing control of it.
Contextual Agentic (CA) is an approach to building AI agents that work inside the context of a real enterprise: its data, its rules, its systems and its people. The team behind it works in enterprise architecture, AI solutions and engineering. We build, test and write about agentic systems, and about how to make them useful while people keep control of authority, evidence and outcomes.
A short note when the book is available, when companion material is released, and when a course opens. Nothing else.
The email list is not open yet, and rather than take an address we cannot yet store, here is what works today: the article feed carries every piece as it is published, collects nothing and asks for nothing. You can also follow the Contextual Agentic Page on LinkedIn (opens in a new tab). Or write to contact@contextualagentic.com and we will add you by hand when the list opens.
No tracking, no address stored, no account.