A preview of the new book by Sho Shimoda — The AI Contact Center Handbook: A Practical Field Guide to Customer Experience in the Age of Agentic AI — now available on Amazon.
What the book is about
The customer contact center is quietly going through the biggest operational shift it has seen in twenty years. The last shift moved conversations from the phone to the chat window. This one is different. This one is not about the channel — it is about who, or what, is doing the work on the other side. Agentic AI does not answer questions and then wait to be told what to do next. It reads the situation, decides what to do, uses the tools it has been given, and completes the outcome. And it is arriving in production contact centers right now, at a pace that most CX leadership teams are underestimating.
This book — The AI Contact Center Handbook — is a practical field guide for the people who are going to have to lead through that shift. It is written for VPs of customer experience, contact center directors, CX operations leaders, and the engineering leaders they partner with. It is not a technology tour and it is not a strategy deck. It is a working manual for how to think, plan, and execute across twelve fronts of the transition — from the definition of "agentic" itself, through workforce redesign, quality and QA in an AI-mediated world, vendor selection, governance, escalation design, cost modeling, and organizational change.
How the book is structured
The book runs twelve chapters organized into five parts, plus six practical appendices. Part 1 defines the landscape. Part 2 works through the front-line operational shifts. Part 3 addresses quality and measurement in AI-mediated conversations. Part 4 covers governance, escalation, and safety. Part 5 covers execution — cost modeling, vendor selection, and change management.
| Part | Focus | Chapters |
|---|---|---|
| Part 1 The Landscape |
What has actually changed, what "agentic" really means, why the operating model must change with it. | Ch 1 – 3 |
| Part 2 The Front Line |
Intent design, tool orchestration, and redesigning the human agent role for exception handling. | Ch 4 – 5 |
| Part 3 Quality & Measurement |
New QA problems that emerge when an agent is doing the talking, and how to measure them. | Ch 6 – 7 |
| Part 4 Governance & Safety |
Escalation design, guardrails, and the risk conversations leaders lose sleep over. | Ch 8 – 9 |
| Part 5 Execution |
Cost modeling, vendor selection, and the change-management work needed to bring the organization along. | Ch 10 – 12 |
A sample — what "agentic" actually means (in plain English)
"Agentic" is a word that spent most of 2024 as a marketing slide and then, sometime in 2025, quietly started meaning something. The reason it started meaning something is that the software behind the word actually changed.
Before the definition, it helps to see the difference in a scenario. A customer contacts a bank because a scheduled bill payment failed.
| Chatbot (2020) | Agent (2026) |
|---|---|
| Asks the customer to describe the problem. | Reads the description and looks up the customer's account. |
| Matches the description against a set of scripts. | Sees the payment failed because the payee changed routing information. |
| Surfaces a static help article — or transfers to a human. | Updates the record, retries the payment, confirms it went through. |
| Result: Customer is now in a queue, waiting for someone else to solve the problem. | Result: Customer is messaged with what happened and what changed. |
In plain English: A chatbot answers. An agent acts. The difference is not the language model. The difference is whether the software is allowed to touch systems, take steps, and change state on the customer's behalf — and whether it can string those steps together toward a goal you gave it in a sentence.
The three ingredients that make software "agentic"
Every genuinely agentic system has three ingredients underneath. Once you can name them, the marketing slides stop being confusing.
A language model that can plan — not just generate text, but decompose a request into steps. When a customer says "cancel my subscription but let me keep access until the end of the billing cycle," a planning-capable model recognizes this is not one intent but two, and that the two must happen in a specific order. Tools the model can actually call — functions like "look up account," "issue refund," "schedule callback," each with a defined input and output. Without tools, the model can only produce sentences; with tools, it can produce outcomes. Memory that spans the conversation — what the customer said three messages ago, what has been tried, what is still outstanding. Memory is the difference between a system that feels like a phone tree and a system that feels like a colleague who has been paying attention.
The one-line takeaway: Agentic AI is a language model plus tools plus memory, wired together so the system can pursue a goal without waiting for step-by-step instructions.
Why this changes the operating model
The interesting consequence is what happens to the human on the other side of the interaction. A chatbot creates a queue: the customer describes, the chatbot deflects, the human eventually inherits everything the chatbot couldn't do. An agent does the reverse. It handles the middle of the work — the account lookup, the routine fix, the confirmation — and hands the human only the exception, with the transcript and the actions already taken delivered at the point of handoff.
Contact center leaders often ask what this means for headcount. The honest answer, worked through in Chapter 4, is that it changes the shape of the workforce more than the total size. The volume of routine work drops sharply. The volume of complex work rises, because the exceptions the agent could not resolve are all now landing on humans at once. The right response is to redesign the human role for exception handling and emotional labor — not to celebrate a headcount reduction that will not deliver.
A concrete example — agentic AI inside Microsoft Teams
The abstract description becomes easier to see with a working example. Consider an enterprise that runs its internal IT and HR support inside Microsoft Teams — where employees already work — rather than in a separate ticketing portal. When an employee messages the support channel asking to reset a distribution list, or to check the status of an equipment order, or to find the current PTO policy, an agentic assistant reads the request, invokes the right internal tools (Active Directory, the procurement system, the HR knowledge base), returns the answer, and — if the action is authorized — takes it.
This is exactly what AB Support — an AI customer service product built by ActionBridge for Microsoft Teams — is designed to do. It lives inside Teams, connects to the systems your employees already use, and handles the middle of the support conversation so your human support team gets the exceptions with full context rather than the routine tickets with none. It is one working example of the pattern this book describes: the model plans, the tools act, and the memory carries the conversation forward. For enterprises whose employees already spend their working day inside Teams, it is the fastest way to see agentic AI operating on a real workflow rather than on a slide.
Who should read this book
| If you are… | The book gives you… |
|---|---|
| A VP of CX or Contact Center Director | A twelve-front framework for planning and leading the transition. |
| A CX Ops or WFM leader | Workforce redesign patterns for the new shape of work. |
| A technology / engineering leader | Shared vocabulary and shared decisions with your CX partner. |
| A consultant, analyst, or investor | A working framework for how to think about this shift. |
| A QA lead, intent designer, or escalation architect | Appendices — glossary, RFP template, QA rubric, playbooks — that earn back the cover price on their own. |