Software and AI
AI agents that use tools and data.
An agent is not just a model that answers: it has instructions, memory, tools connected via API, company knowledge with cited sources and behaviour limits.
AI agent diagram: user, agent with instructions, memory and guardrails, tools connected via API, company knowledge with cited sources, control and evaluation.
Scroll the diagram sideways
Example
One conversation, and what happens behind it.
The agent reads the request, queries the authorised tools, proposes and, only after the user confirms, changes the systems. Every step stays in a log.
Example of a conversation with a bookings assistant and the trace of the agent steps.
Scroll the diagram sideways
What an agent includes
- Instructions and role, with defined tone and limits
- Tools connected via API (calendar, CRM, email, database)
- Company knowledge (RAG): documents, price lists, procedures, with cited sources
- Guardrails: allowed actions, data that must never be exposed
- Evaluation on test cases before and after every change
- Handover to a human operator when needed
Stated limits
- Errors cannot be ruled out: critical steps include human approval
- Answers depend on the quality and currency of the documents provided
- GDPR and AI Act obligations are assessed case by case
Voice agents
The agent on the phone.
Voice agent flow: call, switchboard, speech recognition, AI agent with tools, speech synthesis, transcript and handover to an operator.
Scroll the diagram sideways
The voice agent connects the phone network to speech recognition and synthesis and to an agent with tools. Each call is transcribed and logged under the agreed rules; an operator can take over.