AI USE CASES YOU CAN REUSE.English edition · Individuals, independents, small businesses
GUIDE · 6 CARDS

Agentic AI: how multi-agent systems work in production

Agentic AI refers to systems capable of planning a sequence of actions, calling external tools (APIs, databases, MCP servers) and verifying results before acting — unlike a simple question-and-answer chatbot.

The production stories below show how engineering teams keep these agents safe (restricted permissions, human approval and continuous evaluation).

See also the MCP server guide and create an AI agent and AI agents in company and all use cases.

Production agentic systems and guardrails

  • Triage mobile app crashes with a read-only agentSpotted · Blog technique de ManoMano · OpenAI Agents SDK

    ManoMano runs agents every four hours to review mobile app crashes and match known issues to existing tickets. Application code handles new ticket creation.

  • How Carrefour answers recurring data platform questions with AISpotted · LeMagIT · Agent IA (Google ADK)

    Carrefour built an agent that draws on conversations from its platform engineering team to answer data platform questions the team has handled before. It later extended the agent to other teams.

  • Prepare a calendar access change for the user to confirmSpotted · Blog technique de Doctolib · Alfred (Doctolib)

    Doctolib built Alfred to support healthcare professionals. Its agents interpret the request, look up relevant information and prepare a change for the user to confirm with a click.

  • Delegate client file administration to a team of AI agentsSpotted · France Num (francenum.gouv.fr) · Multi-agent system

    A small retirement advisory firm commissioned a multi-agent system that extracts data from documents, updates the CRM, identifies missing documents and follows up with clients.

  • Prepare merchant outreach with a shared sales agentSpotted · Dust — Alma · Dust

    Alma, a French installment payment provider in around ten European markets, built a Salesforce-connected agent to qualify opportunities and draft outreach tailored to each merchant.

  • Triage technical support tickets and update runbooksSpotted · Dust — Wayflyer · Dust

    Wayflyer, a Dublin fintech that finances growing brands, built two agents. One starts investigating a support ticket when it arrives. The other proposes runbook updates once it is resolved.

FREQUENTLY ASKED QUESTIONS

In brief, with the sources.

What is the difference between generative AI and agentic AI?

Generative AI produces text or images on demand. Agentic AI pairs a model with tools (search, ticket creation, CRM queries, MCP) and a decision loop to complete an end-to-end task under human oversight (Doctolib).

What guardrails are common in agentic AI deployments?

Production lessons from ManoMano and Carrefour highlight the need to restrict automatic write permissions and require human approval on sensitive actions.