Book 2 of 3Forthcoming · 202612 chapters · 3 parts

Prompt Systems & Agent Orchestration.

Engineering Multi-Model AI Workflows

Book 2 of the Full-Stack AI Engineering Series. The application layer. An agent is not a clever prompt. It is a governed actor with a contract, a boundary, and a record.

By Dr. N. Khan · The Full-Stack AI Engineering Series
Set inside Nebula Financial, a fictional regulated fintech · System: AgentMesh

Tell me when it ships → Read the published playbook

Forthcoming · 2026 · the application layer

§ 01Overview

A single clever prompt delights on Monday and embarrasses you on Thursday. The day a stakeholder asks what your agents do, what they are allowed to touch, and what one of them did last Tuesday at three in the afternoon, the team discovers the answer was never built. It was only hoped for.

An agent that demos well and fails in production fails silently, in language fluent enough to be believed. An institution that cannot review what its agents decided, before the decision takes effect, has surrendered control of its own liability without noticing. The problem was never the prompt. The problem is orchestration.

AgentMesh is the catalog of record and the orchestration engine: it turns a pile of prompts into governed agents that plan, act, verify, and submit to human review. Each agent declares a contract, runs inside a boundary, and leaves a record. Orchestration, not cleverness, is the discipline that scales.

§ 02The system · AgentMesh

Five layers, each constraining the next.

Layer 1

Catalog of record

Every agent registered with an identity, a contract, a review tier, lineage, and observability. The five questions a folder of prompts can never answer.

Layer 2

The four-module flow

Planner, executor, verifier, generator. The PEVG unit: separation of concerns made structural, so no single module holds all the power.

Layer 3

The agent graph

Workflows modelled as activity-on-vertex graphs. The workflow in the only form that can be reasoned about, not a diagram beside it.

Layer 4

Tiered human review

Auto-approve, lightweight review, full supervisory review. The review tier rises with the stakes and falls only on a verdict.

Layer 5

Observability envelope

Observable by construction, so the system can reconstruct why it did what it did, before anyone has to ask.

Every registered agent satisfies four obligations before AgentMesh admits it: a contract, a boundary, a record, and a review. An agent does not exist at Nebula until it has declared all four.

§ 03The patterns

The patterns that turn prompts into governed actors.

PEVG

Planner, executor, verifier, generator. The planner decomposes, the executor acts under contract, the verifier judges truth, the generator gives it voice and may never add what was not verified.

Research anchor

The Capability Contract

A small, inspectable, machine-readable declaration of domain, powers, data scope, and tools. Not documentation. The declared boundary the system holds the agent to.

Agent graphs

Workflows as activity-on-vertex graphs: nodes are agents, edges are dependencies, parallelism is read off the structure rather than declared.

Research anchor

Tiered human-in-the-loop

Three review tiers mapped to payment size and regulatory impact. Escalate on a whisper. Downgrade on a verdict.

Research anchor

The Critique Object

Reviewer feedback as first-class, typed data. An edit is data about one answer. A reason is data about a pattern.

Safety rails

Data-only versus action-capable separation, least privilege, PII minimization, and red-teaming at four seams, behind a pre-registration safety review.

§ 04The cost of ungoverned agents

A system that cannot reconstruct why it acted is not an architecture. It is a rumor.

A folder of prompts

A catalog of record

No one can say what an agent did last Tuesday, or what it was allowed to touch.

Identity, contract, review, lineage, observability, on every agent.

An over-privileged agent does, with full authority, what no one intended it to be able to do.

Least privilege, granted by exception, conditioned, and revocable.

A disclosure that is fluent and incorrect, trusted because it is fluent.

A verifier that can say "I do not know" and block before harm.

Reviewers drown in low-risk approvals and rubber-stamp the high-risk ones.

Tiered review that spends human attention where the stakes are.

An irreversible deed, already done, that no downstream check can recall.

A boundary between read-only and state-changing drawn before the act.

§ 05Inside the book

A reference you work from.

Chapters

Parts

PEVG
modules

Reusable
patterns

Part I — From Prompts to an Agent Ecosystem

From Single Smart Prompt to Agent Ecosystem. Why one prompt stops scaling, and the catalog of record that replaces the folder.
Agent Taxonomy and Capabilities in Fintech. Classifying agents by domain, capability, and autonomy, and the contract each must declare.
The Planner, Executor, Verifier, Generator Pattern. The four-module unit, each module's contract, a fraud case worked through all four.

Part II — Trainable, Graph-Structured Orchestration

Training Better Planners from Live Trajectories. Logging the plan, calls, corrections, and outcome as one labelled example.
Agent Graphs. Modelling workflows as activity-on-vertex graphs you can inspect, version, and parallelize.
Dynamic Workflow Refinement. A refiner that reads live metrics and proposes graph edits, then proves them before adopting them.
Agent Registries and Discovery Protocols. Serving humans and agents from one catalog over the Model Context Protocol.

Part III — Human Review, Feedback, and Safety

Human-in-the-Loop as an Orchestration Layer. Three review tiers, per-node defaults, and escalation mapped to risk.
Designing Feedback as First-Class Data. The critique object, a machine-readable schema, fed back into training and compliance.
Safety Rails. Tool boundaries, data minimization, role separation, and red-teaming the seams.
Case Study: Automated Regulatory Disclosure Drafting. Every rail run through one high-stakes workflow, with its audit record.
The AgentMesh Blueprint. The reference architecture, seven patterns and their anti-patterns, and the handoff to Book 3.

§ 06Who it is for

For the engineer who has outgrown the single prompt.

AI engineersApplication developersAgent buildersOrchestration engineersML engineersTechnical leads & architectsTrust & safety engineersPlatform teams

No background in regulation, finance, or enterprise architecture is required; Nebula Financial supplies the regulated pressure. What is assumed is a mindset: comfortable calling a model from code, reading a schema, and reasoning about a contract. Governance is not the tax on building agents. It is the thing that lets you build the thousandth one.

§ 07From the manuscript

An agent that cannot say "I do not know" will eventually say something false with total confidence.

Prompt Systems & Agent Orchestration
Chapter 3 · Draft manuscript

§ 08The series

One discipline, observed from three altitudes.

Three books, one fictional regulated fintech, Nebula Financial, and three systems that are not three products but three faces of one platform, each owning a layer of the stack.

Book 1 · NexusCore LLM Systems in Production Infrastructure The routing and observability gateway that decides which model may answer, under which latency, cost, and risk budget, and records the decision as evidence. View Book 1 →

Book 2 · AgentMesh Prompt Systems & Agent Orchestration Application The catalog of record and orchestration engine that turns a pile of prompts into governed agents that plan, act, verify, and submit to human review. You are here

Book 3 · ThinkFlow DevOps for AI-Native Platforms Operations The AI-augmented internal developer platform that builds, ships, and governs the models, agents, and code the other two layers depend on. View Book 3 →

§ 09About the author

Dr. N. Khan

Dr. N. Khan is an enterprise AI architect and governance advisor with twenty-five years building AI and machine-learning systems at scale. As Principal Architect at iSystematic, he designs the full stack of governed production AI: the LLM infrastructure that routes it, the agents that act on it, the platform that ships it, and the governance that keeps all three defensible.

His practice sits at an unusual intersection of supervisory regulation, quantitative model risk, and enterprise architecture (TOGAF, DMBOK, ISO 27001, SOC 2). He holds a PhD spanning neuro-marketing and computer science, and he is the author of the AI governance Enterprise Playbook.

Dr. Khan writes as a practitioner. His frameworks are built to be used, contested, and adapted, not merely read. He is based in Winnipeg, Canada, with active advisory engagements across MENA. More at nabeelkhan.com.

§ 10Be first to read it

When it ships, you will know first.

The book is in draft. Leave a name and a working email, and you get one note when Book 2 publishes. No list, no noise, no second message.

You are on the list.

One note when Prompt Systems & Agent Orchestration publishes. Until then, the Enterprise Playbook is out now.

An agent is not a product. It is a workload that has not yet met its platform.

Tell me when it ships → Next: Book 3 → Work with the author →

Fin · Book II of III

Five layers, each constraining the next.

The patterns that turn prompts into governed actors.

PEVG

The Capability Contract

Agent graphs

Tiered human-in-the-loop

The Critique Object

Safety rails

A system that cannot reconstruct why it acted is not an architecture. It is a rumor.

A reference you work from.

Part I — From Prompts to an Agent Ecosystem

Part II — Trainable, Graph-Structured Orchestration

Part III — Human Review, Feedback, and Safety

For the engineer who has outgrown the single prompt.

One discipline, observed from three altitudes.

When it ships, you will know first.

Notify me when Book 2 is out

You are on the list.