AI agents » open up a new stage: that of digital assistants capable of planning, searching, writing, sending and even coordinating other AIs.

We’re starting to delegate repetitive tasks to them. No more mile-long prompts, we can ask them: « Sort my e-mails by priority », « Summarize yesterday’s news » or « Identify ten potential prospects », and they come up with a first version. These new digital colleagues still have their moods; sometimes zealous, sometimes distracted, they can run on a loop or miss an instruction. So it’s best to keep an eye on them.

Nevertheless, their uses are multiplying: personal productivity agents (diary, emails), customer service agents (automatic replies and routing), analysis agents (collection and synthesis) or creative agents (texts, images, reports). And each has its own speciality: business agents keep the company running, code agents speed up the work of developers, and data agents extract the trends hidden in the figures.

Here, from the most accessible to the most ambitious, is our selection of AI agents you need to know.

ChatGPT Agent, the digital secretary

Connected to your Gmail and Google Calendar, for example, it can read your messages, pinpoint important exchanges, suggest standard responses and schedule corresponding appointments. In theory, all you have to do is tell it: « Manage my e-mails for the week: prepare urgent replies and block the slots in my calendar. » In practice, the tool is still in beta version (October 2025) and requires manual configuration. The agent begins by scanning the mailbox, classifying messages by priority, writing drafts and proposing them for validation (which must be validated before sending).

Its style? Efficient, impersonal and unadorned. Its limitations: it remains incapable of interpreting innuendo or managing ambiguity (« see you soon? » is not yet a precise date) and the beta version remains prone to bugs.

Relevance AI, the tireless salesman

Relevance AI talks about an « AI workforce »: a digital workforce that performs time-consuming office tasks. It’s a « collection » of agents. Its prospecting agent, for example, acts like a methodical salesperson. You start by defining the profile: « tech companies in Casablanca with more than 50 employees ». The agent searches the Web, locating relevant companies, then enriches each file with public information: contact, site, workforce, sector. He then drafts a customized e-mail. Once validated, he sends the message via your CRM (customer relationship management system: HubSpot, Salesforce, Pipedrive), and records every action. Every morning, he summarizes new leads, e-mails sent and replies received.

Relevance AI doesn’t replace human business acumen, but it does automate everything that gets in the way of it: searching, inputting, sending and follow-up.

Zapier AI, the tireless intelligence agent

Zapier was already the king of workflow automation; with its Zapier AI Agents, it becomes a reader, a watchdog, a personal reporter. The principle: link your information sources (RSS feeds, media sites, newsletters) in just a few clicks and let the agent compile the latest news.

For example, it scans the sites you’ve listed, locates articles containing selected keywords, extracts the essential passages, and drafts a clear digest sent by e-mail or Slack. The result: you receive your press review every morning at 8:30 a.m., clean, sorted and digested.

n8n, monitoring agent and visual automation

n8n transforms scattered tasks into a single logical sequence. Its visual interface connects tools like links: read a source, filter, summarize, send.

In an automated intelligence scenario, n8n explores news feeds, locates relevant articles on AI, for example, has them summarized by a connected model (OpenAI, Anthropic, Mistral, etc.) and sends the whole thing on Slack or by e-mail. The agent acts quietly, at set times, and keeps a log of actions.

Its appeal lies in its freedom: you can adjust the rules, insert a human review, add an internal base. But beware: although powerful, this tool is not « no code » and requires a minimum of technical knowledge.

CrewAI, the invisible team

What if your next collaborator wasn’t a person, but a team of AI agents? That’s the idea behind CrewAI, an open-source project that orchestrates several specialized intelligences, capable of dividing up the workload as in a real organization.

Let’s take the case of a company preparing a call for tenders. A first agent analyzes the specifications, extracting technical constraints andkey dates. A second agent searches internal archives for similar projects and reference data. A third drafts an initial structured response, incorporating relevant examples. Finally, a fourth checks for consistency, clarity and conformity before final validation.

These agents dialogue with each other, correcting each other, sending files to each other and re-sending incomplete tasks. The human only intervenes at the end, to proofread, adjust the tone, or validate the sending. Installation requires technical knowledge (Python, API, workflow management).

LangChain, the brains of customized agents

Where n8n orchestrates actions, LangChain learns to reason. It’s a framework designed for developers: no neophytes please.

The difference is clear. Most tools follow a fixed, step-by-step scenario. LangChain, on the other hand, allows AI to choose its own path. It introduces a real « chain of reasoning »: the agent is given an objective, breaks down the problem, selects the right tools, acts a first time, observes the result, then adjusts its strategy.

A concrete example: an automated analyst in an investment firm. Every morning, he reads the economic reports, extracts the key figures, spots an anomaly or a trend, then looks for the cause in other data. If the results don’t match up, he reruns his research, redoes his calculations and reformulates his conclusions. It’s a process which, through iterations, comes close to human reasoning.

LangChain is not a turnkey product, but an intelligence foundation on which projects such as CrewAI and AutoGPT are built. It is the most complex tool in our selection.

Written in French by Zakaria Choukrallah; Edited in English by AngloMedia Group.