AUTOMATION · Live masterclass
The AI gap is growing. Which side will you be on?
Agents with AI.
Move from using AI to delegating work to AI.
Learn how to design and run an AI agent, under human control.
Next session · November 18, 2026 · 6 h live · 990 €
Follow the complete build of an agent with n8n live: a precise mission, connected tools, human approvals and execution tracking. No programming prerequisites. Leave with the workflow, the templates and the replay to redo it.
n8n
Claude
ChatGPT
Make
Zapier- Microsoft Copilot Studio
- See all tools
Live · 6 h (3 × 2 h) · Intermediate · Replay included · FR or EN
Next session · November 18, 2026 · 6 h live · 990 €
Outcomes
What you will be able to do
- Tell an assistant, a workflow and an agent apart, and know when an agent is justified
- Design an agentic workflow: mission, limits and success criteria
- Connect an agent to tools and data sources
- Automate a multi-step process with n8n
- Set rules, human approvals and handovers to a person
- Identify which tasks an agent can execute, and which ones to keep
- Supervise, control and evaluate an agent: execution history, costs, quality of results
You follow the complete build live. The workflow, templates, guide and replay then let you rebuild it and adapt it to your own mission.
What's at stake
Agents are arriving in work tools. Knowing how to frame them is becoming a skill.
The major vendors are building agents into their platforms. A poorly framed agent is costly, makes mistakes or acts without control. A well-framed agent takes on a repetitive task, under your supervision.
An agent is not a smarter assistant. It is a system that acts. You need to know when to build one, and how to control it.
Assistant, workflow or agent
Assistant
You ask a question, it answers. You act.
Workflow
Fixed steps, defined in advance, run automatically.
Agent
A mission and tools; it chooses the steps, within the limits you set.
Most needs are solved with a workflow. We show you when an agent is really justified.
Example of result
What you will see built
The shared case: an agent that handles incoming requests for a small organization. It is not a sequence of fixed steps: depending on the request, the agent chooses which tools and sources to use.
- Knowledge base (FAQ, terms, fictitious prices)
- Customer history (fictitious file)
- Calendar
- Clarification request to the customer
- Handover to a person
Human approval before any action
Its mission
Understand the request, gather the necessary information with the right tools, then submit a proposed action for human approval. Nothing is sent and no external action is taken without approval.
Its limits
- List of authorized tools
- Maximum number of steps
- Mandatory handover to a person for out-of-scope cases
Three requests, three paths
Request 1
A simple question about cancellation terms
Tools chosen
Why
The answer is in the terms; no customer data is needed.
Proposal submitted
A proposed reply
Request 2
An appointment request from an existing customer
Tools chosen
Why
It needs the customer's file, then a free slot.
Proposal submitted
A proposed time slot
Request 3
An ambiguous request, partly out of scope
Tools chosen
Why
The agent cannot decide alone: it asks for clarification, or hands the case over.
Proposal submitted
A question to the customer, or a case handed over
Depending on the request, the agent chooses its tools. A person approves before each action.
Pedagogical case — fictitious data.
Other possible missions
- Prepare a weekly watch on a topic and send it for review
- Qualify sales requests and suggest a priority
- Prepare a briefing before a meeting from defined sources
The method
The essential choices for a first useful agent
Our selection comes from the tools we use and evaluate on our projects. At each step, you see the decision made and the reason for it.
- 1Which mission
- 2Which limits
- 3Which tools
- 4Which controls
The tools
The AI tools you will explore & learn to use
For each tool: what it is for, its strengths, its limits and when to choose it. The goal is not to master every platform, but to know which one fits your mission. Each one has its sheet in the H5 Guide.
Build the agent
Automation and agents, from the most controllable to the most accessible.
- n8n
AI agent, tools, memory, human approvals, execution history
- Make
Agents inside a visual automation
- Zapier
Agents connected to a very large number of apps
Frame and reason
The mission sheet, the instructions and the model that chooses the actions.
- Claude
Write the agent's mission sheet and instructions
- ChatGPT
ChatGPT's built-in agent, to compare with a custom-built one
- An AI model
Reason and choose the actions
In the enterprise
The criteria to review with your IT department: data access, agent permissions, traceability.
- Microsoft Copilot Studio
Agents in the Microsoft 365 environment
- Gemini Enterprise
Agents in the Google environment
- Dust
Agents connected to internal tools, European vendor
Agentic frameworks
To go further in code, with a technical team.
- OpenAI Agents SDK
Build agents in code
- Claude Agent SDK
Build agents in code
LangGraphOrchestrate several agents in code
CrewAIOrchestrate several agents in code
- Langfuse
Track and evaluate executions
- Claude Code
Development agent
Standard
Model Context Protocol (MCP)
The standard that lets you connect an agent to your tools. Every connection is an access to control.
The names and logos mentioned belong to their respective owners. H5 AI Academy is neither affiliated with nor sponsored by these vendors.
Program
A complete agent, built live in three steps
| Session | What you see built | What you take away |
|---|---|---|
| 1 — Understand and frame | Assistant, workflow or agent: how to choose. Choosing the mission, its limits and its success criteria. Writing the mission sheet and instructions with Claude. | Mission sheet template, instructions template, "do you need an agent?" grid |
| 2 — Build | Building the agent in n8n: trigger, AI model, available tools, knowledge base, limits. Running it on three different requests, and reading the history to see which tools the agent chose. | The agent's n8n workflow, to import |
| 3 — Make it reliable | Tests on normal, ambiguous and out-of-scope cases. Adding human approvals and handovers to a person. Tracking executions and costs. Short, timed demonstration of the founder's stack with Claude Code. | Control checklist, testing grid, cost tracking sheet |
1 — Understand and frame
What you see built — Assistant, workflow or agent: how to choose. Choosing the mission, its limits and its success criteria. Writing the mission sheet and instructions with Claude.
What you take away — Mission sheet template, instructions template, "do you need an agent?" grid
2 — Build
What you see built — Building the agent in n8n: trigger, AI model, available tools, knowledge base, limits. Running it on three different requests, and reading the history to see which tools the agent chose.
What you take away — The agent's n8n workflow, to import
3 — Make it reliable
What you see built — Tests on normal, ambiguous and out-of-scope cases. Adding human approvals and handovers to a person. Tracking executions and costs. Short, timed demonstration of the founder's stack with Claude Code.
What you take away — Control checklist, testing grid, cost tracking sheet
A format designed for following along
One shared case, built end to end by the instructor. There is nothing to install or configure to follow. The replay, the workflow, the templates and the guide then let you redo each step and adapt it to your own mission.
Inspired by real situations from the H5 ecosystem.
Next session · November 18, 2026 · 6 h live · 990 €
What you take away
The H5 Guide to AI agents
One sheet per tool: what it is for, when to choose it, its limits, and the accounts and plans needed to redo it on your own.
- The agent's n8n workflow, to import, without credentials, with its setup instructions and fictitious data
- The templates: mission sheet, instructions, "do you need an agent?" grid
- The control checklist: data, access rights, human approvals, costs, execution log
- The testing grid
- The replay and the certificate
Who it is for
Project and operations managers
Know which tasks to hand to an agent, and which to keep.
Entrepreneurs and small-business leaders
Automate a repetitive task without losing control.
Freelancers and consultants
Know how to frame and price an agent for clients.
Business teams in companies
Understand the agents built into their tools and the criteria to review with IT.
No programming prerequisites. Having followed Work with AI is recommended.
The questions to examine before handing a task to an agent: data, access rights, human approvals, traceability.
Train my teamYour instructor
Haron Hasni
Founder of H5 AI Academy
29+ years in technology, data and business applications. Projects for large groups. AI companies launched.
1,200+ people trained · 300+ projects · 50+ organizations
“A useful agent is first a well-framed agent. I show you how I decide what it can do, and what it won't do.”
Frequently asked questions
Do I need to know how to code?
No. You follow the build, and every choice is explained.
Will I build during the course?
No. You follow the complete build live. The replay, the workflow, the templates and the guide let you redo it.
How is it different from Build with AI?
Build creates an application prototype. Agents with AI designs a system that carries out a mission with tools, under human control.
Can an agent act without me?
In the demonstration, no sensitive action is executed without human approval. We show you how to decide the level of autonomy, mission by mission.
How much does it cost to run an agent?
It depends on the model, the volume and the hosting. We show you how to estimate and track these costs.
Do I need to create accounts or pay for subscriptions?
Not to follow the course. To redo it on your own, the guide lists the accounts needed and each tool's plans.
Can I connect an agent to confidential data?
We show you the questions to ask and the access rights to check before any connection.
What if the course isn't right for me?
30-day guarantee. Refund Policy
Before handing a task to an agent, learn how to frame it. Give yourself 6 hours.
Next session · November 18, 2026 · 990 € · Guide · Workflow · Templates · Replay · Certificate
November 18, 2026
990 €