AI assistants and agents
AI agents training: from design to production
Building an agent has become easy; running it properly has not. Two levels to learn how to design an agent on a real case, test it, integrate it with your tools, then orchestrate several agents and supervise them.
Advanced level · Automating your tasks and integrating an AI agent
- Duration
- 3 days (21 hours)
- Level
- Advanced
- Format
- On-site / Remote
- Price excl. VAT
- €4,500 excl. VAT in-house, per group€1,500 excl. VAT open session, per participant
Expert level · Building and orchestrating AI agents
- Duration
- 4 days (28 hours)
- Level
- Expert
- Format
- On-site / Remote
- Price excl. VAT
- €6,000 excl. VAT in-house, per group€2,000 excl. VAT open session, per participant
Why this training
An agent acts in your systems: so do its mistakes
A chatbot error is a wrong answer on screen; an agent error is an action in a system: a file changed, an email sent, an order cancelled. The hard part is not assembling an agent, it is deciding whether the case justifies one, proving that it works, limiting its rights and being able to stop it.
Cybersecurity authorities agree. In April 2026, CERT-FR recommended, for agentic assistants installed on workstations, mandatory human approval whenever an action with side effects is considered, and tools limited to what is strictly necessary.
The undocumented agent
Built by one person, with instructions nobody has reviewed, it becomes unmanageable when its creator changes role or the model evolves.
Overly broad access
Connecting the agent with your own credentials is the fastest route. An agent that was only meant to read the CRM can then modify it.
No tests
Without a test set, every instruction tweak or model change is a gamble. Testing is the most neglected skill, and the first one the training builds.
Audience and prerequisites
Two levels, depending on your role
The first level is for those who own a process, the second for those who build and run agents. Security (access, secrets, logging) is best involved from the scoping stage.
- Advanced level (3 days): business leads and operations managers who want to automate a process
- Expert level (4 days): technical teams who develop, orchestrate and industrialise agents
- In-house, a business, operations and technical trio working on a real company case
Advanced level
Automating your tasks and integrating an AI agent
- Prerequisites
- Command of the AI agent concept
- Level
- Advanced
Expert level
Building and orchestrating AI agents
- Prerequisites
- Comfortable with AI agent concepts and experienced in configuring agents
- Level
- Expert
Objectives
What participants can do at the end of the pathway
- 01Decide whether a case justifies an agent or conventional automation
- 02Design an agent on a real use case: model, tools, instructions
- 03Build a test set and evaluate the agent before opening access
- 04Integrate the agent with existing tools, with limited rights and human approval of sensitive actions
- 05Orchestrate several agents when justified, then supervise and industrialise them (expert level)
Programme
Two levels, ten modules
Three days for the first level, four for the second, on-site or remote. In-house, the programme adapts to the company's use case and tools.
Advanced level · 3 days (21 hours)
Automating your tasks and integrating an AI agent
Module 1
Benefits and limits of AI agents
Workflow or agent: when an agent is the right answer, when conventional automation is enough, and what an agent costs in latency and model calls.
Module 2
Agent design platforms
Overview and selection criteria for a platform: available models, connectors, rights management, data hosting.
Module 3
Designing an agent for your use cases
Model, tools, instructions: start from existing procedures and the exceptions only the business knows. One agent, one goal.
Module 4
Testing methodology
A set of cases drawn from real situations, automated checks and human review, regression tests before every change.
Module 5
Integration into existing workflows
Connecting to company tools, including through the MCP protocol, the risk level of each action, human approval of sensitive actions.
Expert level · 4 days (28 hours)
Building and orchestrating AI agents
Module 1
Designing and developing an AI agent
An agent's architecture, choice of model, definition of tools and their parameters to make mistakes hard to make.
Module 2
Configuring and testing its capabilities
Instructions, error handling, automated evaluations and review of execution traces.
Module 3
Extending the agent's capabilities
Access to company documents, new tools, context management: every added capability is tested and bounded.
Module 4
Orchestrating several AI agents
Orchestrator and sub-agents, one agent evaluating another: when to split, and what each additional agent costs.
Module 5
Industrialisation and monitoring
Traceability, escalation thresholds to a human, cost tracking, shutdown procedure: running an agent as a production service.
Practical details
Duration, format and prices
Advanced level · Automating your tasks and integrating an AI agent
- Duration
- 3 days (21 hours)
- Level
- Advanced
- Format
- On-site / Remote
- In-house
- €4,500 excl. VATfor a group from the same company
- Open session
- €1,500 excl. VATper participant, inter-company session
Expert level · Building and orchestrating AI agents
- Duration
- 4 days (28 hours)
- Level
- Expert
- Format
- On-site / Remote
- In-house
- €6,000 excl. VATfor a group from the same company
- Open session
- €2,000 excl. VATper participant, inter-company session
In-house, the pathway starts from a real company case and leaves verifiable deliverables: a documented case with a named owner, a first test set, the list of the agent's tools and their rights.
All prices are exclusive of VAT.
Funding
Getting the training funded
JustAI's Qualiopi certification gives access to funding from your French skills operator (OPCO). We provide the quote, the detailed programme and the training agreement: the documents your OPCO expects.
OPCO
Your skills operator (Atlas, Opco 2i, Akto, Afdas…) can cover all or part of the training, in-house or in an open session, under the skills development plan. Each OPCO applies its own rules and ceilings.
Payment: what changes on 1 October 2026
At several OPCOs, including Atlas, AKTO and Opco EP, direct payment of the training provider becomes the exception: the company pays the invoice, then claims reimbursement. Check your OPCO's rule before committing to the training.
CPF
This training is not certified, so it is not eligible for the CPF (France's personal training account). The CPF can fund certified training, such as our 4-day programme preparing for the RS No. 6891 certification on producing and reviewing professional content with generative AI, intended for content-production roles and subject to the applicable cap.
We help you put the funding application together: it is part of the service.
OPCO: what changes on 1 October 2026 (in French)Quality
A certified training provider, an engineering firm
Qualiopi Certification
Certified for professional training actions. Certificate valid from 08/07/2024 to 08/06/2027.
View the certificateWe practise what we teach
JustAI is an AI consulting and engineering firm, based in Paris since 2020. We run our own GPU servers in France, serve open-weight models through our own inference APIs and ship custom software to production. Our training draws on that practice.
Our infrastructureFAQ
Your questions about this training
Which level should we choose?
"Automating your tasks and integrating an AI agent" (3 days, advanced level) suits business leads and operations managers who already understand what an AI agent is. "Building and orchestrating AI agents" (4 days, expert level) is for people who have already configured agents. To discover the topic, the one-day course "Building an AI assistant: why and how?" is a good starting point.
Do participants need to code?
Not for the advanced level, whose prerequisite is a command of the AI agent concept: design platforms have made building an agent accessible to non-developers. The expert level requires experience configuring agents.
Which platforms do you use?
In-house, the tools your company has chosen, on a real case. We teach principles (rights, tests, traceability) rather than the clicks of one tool version, which age quickly.
What should we expect at the end?
Verifiable outputs: a documented use case with a named owner, a first test set, the list of the agent's tools and their rights, and the rule that triggers human approval.
What are the prices and how can it be funded?
Advanced level: €4,500 excl. VAT in-house, €1,500 excl. VAT per participant in an open session. Expert level: €6,000 excl. VAT in-house, €2,000 excl. VAT per participant. Both can be funded by your OPCO under the skills development plan; they are not eligible for the CPF.
Related training
Build the pathway
2 days (14 hours) · Intermediate
Claude Code training for developers
See the training1 day (7 hours) · Advanced
Prompt engineering training
See the training1 day (7 hours) · Intermediate
Building an AI assistant: why and how?
See the training3 days (21 hours) · Advanced
The technical stakes of AI for IT departments
See the trainingFurther reading
Related articles
Our articles are published in French.
Outils & technologies
Agent IA, orchestrateur, chatbot : comprendre les différences pour bien choisir
Read the articleOutils & technologies
MCP (Model Context Protocol) : le standard qui transforme les agents IA d'entreprise
Read the articleOutils & technologies