AI agents and business assistants

Designing AI assistants and agents for business

Move from chat to assistants grounded in your documents and agents that act in your tools, with safeguards.

Request a quote
Level
Advanced
Indicative duration
21 h
Format
Online · On-site
Language
French
Price
On quotation

Learning objectives

  • Tell apart a custom assistant, a document-grounded assistant (RAG) and an agent, and choose the right level for a need.
  • Write robust system instructions and structure a usable knowledge base.
  • Prototype an agent that calls tools through n8n or a model API.
  • Evaluate answer quality with a test set and explicit criteria.
  • Put safeguards in place: scope of action, human approval, logging, cost control.
  • Write a realistic specification for an assistant or agent project.

By the end of this course, you will be able to design, prototype and evaluate an AI assistant that answers from your own documents, as well as an agent that carries out actions in your tools (CRM, email, spreadsheets) under human supervision.

We draw a clear line between levels of ambition: a custom assistant (custom GPT, Claude project, Gem), an assistant backed by a document base (retrieval-augmented generation, or RAG), and an agent that calls tools and chains steps. For each, you learn when it makes sense, what it costs and what can go wrong.

The course focuses on responsible deployment: test sets, measuring answer quality, restricting permitted actions, human approval of sensitive operations, logging, controlling API consumption and the confidentiality of data sent to model providers. The final project is to design an assistant or agent for a use case in your organisation, with its specification and evaluation protocol.

Who it is for

  • Digital project managers, innovation or transformation leads.
  • Technical profiles and developers who want to integrate AI models.
  • Consultants and operations managers running automation projects.
  • SME leaders who want an in-depth understanding of what an agent can and cannot do.

Prerequisites

  • Regular use of at least one AI assistant (ChatGPT, Claude, Gemini or Copilot).
  • Having built a simple workflow in an automation tool (n8n, Make or Zapier), or completed our process automation course.
  • Basic understanding of APIs and JSON; programming is a plus but not required.

Skills you will gain

  • System prompt design
  • Knowledge base structuring
  • Retrieval-augmented generation (RAG)
  • Tool calling and orchestration
  • Answer quality evaluation
  • Agent security and safeguards
  • API cost estimation
  • Writing AI project specifications

Programme

  1. Overview and architecture choices

    3 h
    • Assistant, document-grounded assistant, agent: working definitions
    • Relevant use cases and ones to avoid
    • Consumer plans, business plans and APIs: confidentiality differences
    • Estimating usage cost: tokens, volume, model choice

  2. Custom assistants

    4 h
    • Writing system instructions: role, scope, tone, refusals
    • Preparing reference documents: structure, updates, access rights
    • Custom GPTs, Claude projects, Gems: possibilities and limits
    • Testing an assistant against hard or out-of-scope questions

  3. Assistants grounded in your documents (RAG)

    4 h
    • How retrieval works: chunking, indexing, retrieval
    • Source quality: outdated documents, contradictions, rights
    • Citing sources in answers and handling "no answer"
    • Prototype with a no-code tool or n8n

  4. Agents and tool calling

    4 h
    • Giving a model tools: read, write, send
    • Orchestrating an agent in n8n
    • Limiting scope: least privilege, allow-lists of actions
    • Human approval before irreversible actions

  5. Evaluation, security and operations

    3 h
    • Building a test set and evaluation criteria
    • Risks: hallucinations, prompt injection, data leakage
    • Logging, monitoring and continuous improvement
    • Personal data: principles of Tunisian Organic Law No. 2004-63 and provider contracts

  6. Project: from need to specification

    3 h
    • Choosing a use case from your organisation
    • Writing the specification and success criteria
    • Prototype and first tests
    • Presentation and implementation plan

Exercises and projects

  • Write and test the system instructions for a customer FAQ assistant.
  • Prepare a document base (internal procedures or product sheets) and measure the quality of the answers.
  • Build an n8n agent that reads an incoming request, classifies it and drafts a reply for approval.
  • Write a set of 20 test questions, including traps, and score the answers against defined criteria.
  • Final project: full specification of an assistant or agent for your organisation.

Upcoming sessions

No public session is scheduled yet. Leave a request: we will let you know as soon as a date is set, or organise a session for your team.

How to enrol

How to enrol

  1. Fill in the enrolment request form on this page (name, email, phone, current role and expectations).
  2. We reply by email to check that the course matches your level and to share practical details.
  3. Scheduled sessions are announced on this page as soon as they are planned; if no date is shown, your request is recorded and you will be notified when the next session opens.
  4. Your enrolment is final once confirmed in writing by email.

Companies: to enrol several employees or arrange a dedicated in-house session, mention it in the form and we will send you a quote.

Request a quote

Context, goals, tools in use, timeline.