Mastering AI Agents in the Workplace is a hands-on course in using ChatGPT and Claude as AI agents for administrative work and the IT service desk, from projects, skills and scheduled tasks to ticket triage and knowledge base articles. It suits IT support staff, administrative teams and AI champions, and you leave with a pilot plan for your own unit.
AI tools at work are moving from chatbots that answer one question at a time to AI agents that plan multi-step work, use connected files and apps, produce documents, spreadsheets and reports, and run repeating jobs on a schedule, in both ChatGPT and Claude. Yet most people still use AI for one-off questions and answers, so the productivity gains stay out of reach, especially in administrative work full of repetitive steps and on IT service desks that receive, sort and answer a large volume of requests every day.
This course explains how AI agents work and then puts ChatGPT and Claude to work on real office tasks, using projects, skills, plugins, connectors, ChatGPT Work and scheduled tasks. It then builds an AI-assisted IT service desk, from intake, ticket triage and prioritisation to knowledge base articles and draft replies, and shows how MCP and n8n can connect automated workflows, all within a data security framework. Everything is built around a fictional organisation, NovaWorks, and the course ends with a 30-60-90 day pilot plan for each learner's own unit. (2 days, 6 hours per day, 12 hours in total, Beginner to Intermediate level.)
Course Details
This is a mixed IT and administrative course built around an AI-assisted service desk using ChatGPT and Claude. It runs for 2 days, 6 hours per day (12 hours in total, 09:00-16:00), as lectures with labs and workshops built on a fictional organisation, NovaWorks, that carries through both days. Beginner to Intermediate level. Day 1 builds an admin assistant kit for office work, and Day 2 assembles the NovaDesk AI service desk and a 30-60-90 day pilot plan. Everything is no-code; MCP and n8n are explained and demonstrated without coding or server setup, while in-depth ITSM configuration and building agents with
an API or SDK are out of scope. It differs from AIOps and AI for IT Service Management, which takes the IT operations view, AI Agents for Business, which covers AI agents for business work in general, and AI for Workforce, which focuses on everyday work for all staff. Learners use their own or their organisation's ChatGPT and Claude accounts; the institute does not provide accounts, and the instructor demonstrates paid-plan features for learners on free plans. No real company data is needed, as all workshop data is simulated. Menu names and features of AI tools change often, so the content follows the latest version on the training day.
Section 1: From Chatbot to AI Agent
Chatbots, AI assistants and AI agents, and the levels of autonomy
The parts of an agent: model, instructions, tools, memory and the agent loop (plan, act, check)
Human-in-the-loop: where AI must ask for approval first, and where people must review
Lab: run the same task in ChatGPT and Claude and compare chat with agent-style work
Section 2: Lab: Prompt and Context Engineering for Work
A prompt structure for organisational work: role, task, context, constraints and output format
Consistent results with few-shot examples, tables or JSON, and letting AI ask when information is missing
Reduce hallucination by requiring sources, and mix Thai and English in one document
Lab: build prompt templates for email, meeting summaries and reports
Section 3: Lab: A Knowledge Workspace with Projects
Create projects in ChatGPT and Claude with project-level instructions
Upload policies, SOPs and forms so AI answers staff questions with references
Choose suitable documents, and know what should never go into AI
Create documents, spreadsheets and slides from project data
Section 4: Skills, Plugins and Connectors for Administrative Work
Skills: the idea, turning them on and writing a skill from a real procedure (SKILL.md)
Plugins in ChatGPT that bundle skills and apps, and moving from custom GPTs as they are retired
Plugins and connectors in Claude for email, calendar and drive, with care over permissions
Lab: build a skill that drafts formal letters, memos and emails in two languages
Section 5: Delegating Multi-step Work and Scheduled Tasks
Delegate multi-step work with ChatGPT Work and Claude: research, analyse and deliver a file
Brief like a manager: goal, expected output, scope and acceptance criteria
Scheduled tasks: have AI produce reports or summaries on a regular cycle
Track progress, step in along the way and check the work before using it
Section 6: Workshop: The NovaWorks Admin Assistant Kit
Build a knowledge project from simulated policies and SOPs that answers staff with references
Use the skill to draft formal letters and emails in two languages from NovaWorks requests
Delegate a scheduled weekly summary report of requests to an AI agent
Check the output against an accuracy and data safety checklist
Section 7: AI in IT Service Management
Where AI helps a service desk: intake, triage, knowledge deflection, resolution and escalation
ServiceNow Now Assist, Rovo in Jira Service Management and Freddy AI Agent in Freshservice
Use ChatGPT and Claude alongside an existing ticketing system without replacing it
Extend the idea to internal administrative services such as room booking, purchasing and HR
Section 8: Lab: Ticket Triage and Prioritisation with AI
Classify issues and extract key details (requester, device, symptoms) from free-text tickets
Prioritise with an impact/urgency matrix and SLA rules
Spot duplicate tickets and clusters of related issues (patterns or major incidents)
Lab: triage simulated NovaWorks tickets and export the result as CSV or JSON
Section 9: Lab: Knowledge Base and Drafting Replies
Turn resolved tickets into knowledge base articles, FAQs and basic troubleshooting guides
Draft replies in Thai and English pitched at the requester's level of knowledge
Write escalation summaries so the next support tier can pick up the case at once
Lab: build a standard skill for KB articles and replies so the whole team works the same way
Section 10: Connecting AI Agents to Your Systems: Connectors, MCP and n8n
The Model Context Protocol (MCP): tools, resources and prompts that systems expose to AI
Connectors and apps in Claude and ChatGPT for Slack, Google Drive, Microsoft 365 and other systems
n8n demo: the AI Agent node with the MCP Client Tool, and exposing a workflow through the MCP Server Trigger
Example workflow: a form or email report, AI triage, a logged ticket and a chat alert to the team
Section 11: AI Governance, Security and Measuring Value
Data classification, personal data (PDPA) and what must never go into AI
Prompt injection, excessive permissions, least privilege and audit trails in organisation plans
Metrics: deflection rate, first contact resolution, MTTR, time saved and user satisfaction
Lab: draft an AI usage policy for your unit
Section 12: Master Workshop: The NovaDesk AI Service Desk
In mixed IT and admin groups, assemble NovaDesk from ticket intake through triage and prioritisation
Find answers in the knowledge base, draft replies to users and write escalation summaries
Produce a weekly service performance report
Present a 30-60-90 day pilot plan with metrics and governance for your own unit
No public rounds are open right now. Join the waiting list and we will contact you first when the next round opens, or ask us on LINE. Or call 02-570-8449 or 088-807-9770
AIOps and AI for IT Service Management looks at ITSM from the IT operations side, AI Agents for Business covers AI agents for business work in general, and AI for Workforce focuses on everyday AI use for all staff. This course is the mixed IT and administrative course built around an AI-assisted service desk using ChatGPT and Claude. Day 1 builds a toolkit for office work, and Day 2 turns it into a service desk where AI helps triage tickets, draft replies and build a knowledge base, ending with a 30-60-90 day pilot plan.
The institute does not provide accounts or subscriptions. You use your own or your organisation's ChatGPT and Claude accounts. A free account covers most workshops, such as chat and document work, and Claude's free plan also includes skills and connectors with limits. Features such as ChatGPT Work, scheduled tasks and plugins need a paid plan; if you only have a free account, the instructor demonstrates them and adapts the workshop for you. Organisation accounts may need an administrator to enable the features used in class beforehand.
Yes. The course is no-code, and every workshop uses the normal ChatGPT and Claude interfaces. MCP and n8n are explained and demonstrated so you can see how they work, and IT learners who are interested can try them further at their own pace.
No. All workshops use simulated data from a fictional organisation, NovaWorks, including policies, SOPs, request forms and IT tickets. If you want to try your own work, remove confidential and personal data first, which is the same principle the course teaches on data classification.
Bring a laptop with internet access and an up-to-date browser such as Google Chrome or Microsoft Edge. Before the training day, check that you can reach chatgpt.com and claude.ai and sign in to your accounts. If you want to try features that work with files on your computer, install the ChatGPT and Claude desktop apps in advance.
Both tools change often and roll out features by plan and account. For example, OpenAI has announced that custom GPTs are being retired in favour of plugins, and Anthropic merged Cowork into Claude chat from September 2026, so some accounts may see different menu names or screens. The instructor updates the content and workshops to the latest version on the training day, and the core ideas, such as delegation, human-in-the-loop and governance, still apply when the interface changes.
As systems grow to many services, IT operations and service desk teams run into the same problems. One fault sets off dozens of alerts at once, tickets arrive without the right category, the same issue gets solved again because nobody can find the last fix,…
Organizations have mostly used AI in a question-and-answer style: a person asks, AI answers, and the person then does the follow-up work every time. Today AI has moved beyond an assistant waiting for commands to an AI agent that takes a goal in human language…
Many employees already know about AI and have asked it the odd question, yet it has not become part of their everyday work. They are unsure which tasks to start with, what data they may or may not enter, and how far the output can be trusted, so AI gets used…
