Merci d'avoir envoyé votre demande ! Un membre de notre équipe vous contactera sous peu.
Merci d'avoir envoyé votre réservation ! Un membre de notre équipe vous contactera sous peu.
Durée 14 heures
Plan du cours
- Day 1: Planning agents and building the first one
- From assistant to agent
- Chat, assistant, agent and automation compared
- What agents can do reliably today
- Examples from administration, purchasing, HR and customer service
- Exercise: Classify three own workflows by whether chat, assistant or agent fits.
- The right task
- Criteria such as frequency, clear steps and a verifiable result
- Describing input, processing, output and control point
- Defining success criteria
- Exercise: Fill in an agent profile for the own workflow.
- Building blocks of an agent
- Goal and instructions
- Knowledge from documents and file storage
- Tools such as mail, calendar, tables and web
- Limits and escalation
- Exercise: Write instructions and limits for the own agent.
- The first agent
- Choosing a tool based on approval in the company
- Agent in Copilot Agent Builder, as a custom GPT, a Claude project with a skill or a Langdock agent
- Testing with three real cases
- Exercise: Build the first agent and test it on three cases.
- Outcome: Each participant has an agent profile and a first, tested agent for an own workflow.
- From assistant to agent
- Day 2: Agents with tools, workflows and control
- Agents that act
- From answering to executing
- Connectors to mail, file storage, CRM and tables
- Agents in the browser as an outlook
- Exercise: Connect the agent from day 1 to a tool, for example file storage or a table.
- Agents in workflows
- Triggers such as a new email, a new form or a fixed time
- AI agent step in n8n or Make
- Storing results in tables, tickets or a mailbox
- Exercise: Build a workflow in which an agent classifies incoming requests and stores a draft.
- Human control
- Approval points before sending, booking or deleting
- Log of the agent's steps
- What happens when the agent is unsure
- Exercise: Build an approval via email or Teams into the workflow.
- Testing and operating
- Test cases with expected results and failure cases
- Costs, usage limits and maintenance
- Responsibilities and documentation
- Exercise: Run through five test cases and write short operating documentation.
- Presentation and next steps
- Showing results in the group
- Selecting the next agent
- Clarifying responsibilities for maintenance and further development
- Exercise: Present the own agent in three minutes.
- Outcome: Each participant has an agent with a tool, an approval point, test cases and short operating documentation.
- Agents that act
Pré requis
- Access to at least one agent tool in the company (Copilot Agent Builder, ChatGPT, Claude, Langdock or n8n), and a recurring workflow from the own area.
- No programming skills required.
Audience
- Specialist users, key users and process owners who want to hand over recurring work to AI agents.
Nos clients témoignent (1)
exemples de la vie réelle
Maria - Fundacja PTA
Formation - Mastering Make: Advanced Workflow Automation and Optimization
Traduction automatique
