English pilot — the directory is mainly in French for now.
You hear “AI agent” everywhere, often as a label for any piece of software with a bit of AI in it. This guide puts it in plain words: what an agent really is, how it differs from a chatbot and from classic automation, and what it is actually good for in real life.
An AI agent is software you give a goal — not just a question — and that organizes itself to reach it: it looks up information, uses tools (your inbox, your calendar, a browser, a spreadsheet), chains several steps together and delivers a result. Where a classic assistant waits for your next sentence, the agent works between two of your interventions.
A chatbot (like the chat window on a website) answers what you ask, one reply at a time. It does nothing between messages and generally has no access to any of your tools.
An automation (an inbox rule, a Make or Zapier scenario) always runs the same sequence of steps, in the same order, when a specific event happens. Effective, but rigid: if the situation falls outside the script, it stops or gets it wrong.
An agent sits between the two: you give it a goal and limits, it picks the steps itself and can adapt when something unexpected happens — for example by rereading its own answer, trying another tool, or asking you to approve before going on. That degree of autonomy is what defines it, and it is also why it needs guardrails (see our guide Risks and guardrails, in French).
Every profile in this directory starts from a real problem. Three examples to feel the difference:
— Sort your email and draft the replies while you do something else: Lindy reads incoming messages, files them away and prepares drafts.
— Post to your social accounts without having to think about it every morning: Metricool prepares and schedules your posts.
— Carry out an entire web chore — comparing, filling in forms, booking: ChatGPT (agent mode) browses and acts on websites on your behalf, under your supervision.
Is this really new? The idea is old, but the combination is recent: language models that are reliable enough, standardized ways to plug them into your software (see the glossary, in French, MCP entry) and prices that have become affordable. The current wave dates from 2024–2026.
Is it only for developers? No. Most profiles in this directory are rated “Easy”: finished products you can set up in a few clicks. The Advanced agents category, on the other hand, is aimed at people who build their own agents.
Will an agent replace my job? Today’s agents excel at tasks that are repetitive, bounded and easy to check — sorting, data entry, first drafts, monitoring. They remain poor judges: anything that takes context, accountability or a human touch stays human, and every profile in the directory spells out the guardrails to put in place.
How much does it cost? From $0 (open source, free plans) to a few tens of dollars a month for most uses, with a few special cases billed by usage. The honest breakdown is in the guide How much does an AI agent cost?.
Browse the directory’s 111 agent profiles →