FarolAI
CONCEPTS AND TIPS · 5 min read

What are AI agents and how can you use them at work?

An artificial intelligence agent is a system configured to help achieve a goal. Instead of explaining your workflow from scratch in every request, you define a role, instructions and boundaries that your team can reuse.

What is an AI agent?

An agent combines an AI model with instructions and, when available, knowledge and tools to perform tasks. Depending on the implementation, it may simply respond to requests or carry out several steps using tools.

In FarolAI, you configure agents with a name, command, instructions, skills and permissions. Requests can come from the web or WhatsApp and are recorded in the history. Do not assume an agent works continuously on its own or has access to all your systems.

How is an agent different from a chatbot?

A chatbot is a conversational interface. An agent is a system configured around a goal, which may use context and tools to help achieve it. The concepts can overlap: an agent can also operate through a chat.

In practice, check which instructions it follows, which data it can access and which actions the implementation supports. A sales agent may draft a proposal without being authorized to send it or record a sale.

Where agents can help your team

  • Marketing: draft captions, campaign ideas and message variations.
  • Sales: prepare discovery questions and proposals using the terms you provide.
  • Internal work: organize notes, summarize supplied text and prepare checklists.
  • Support: suggest replies based on provided guidance for your team to review.

Start with one small, repetitive task. Choose an output you can evaluate, such as a short reply that must include specific information.

How to configure a useful agent

  1. Define its role: what problem does it solve, and for whom?
  2. Provide context: relevant products, audience and processes, without unnecessary information.
  3. Describe the output: format, language, length and tone.
  4. Set boundaries: what it must not invent and when to ask questions.
  5. Control access: make the agent and its connections available to the right people.
  6. Test and refine: compare responses to typical requests and requests with missing information.

Use skills to reuse instructions across agents. An instruction-based skill is not the same as an integration or an automatically executed script.

How to write better requests

A useful request includes the task, context and expected output. Compare “write a post” with:

/marketing Write three captions of up to 60 words introducing our new T-shirt collection. Audience: local customers. Tone: friendly and direct. End with an invitation to visit the store. Do not invent prices or discounts.

Include the necessary context in each new request. Do not rely on permanent memory of earlier conversations. If the result misses the goal, explain what needs to change.

Understand the limits

Agents can make mistakes or produce unsupported information. Check names, numbers, deadlines and commercial terms before using a response. Ask the agent to flag uncertainty and request missing information.

A connection must be configured and support the requested task. Asking “send this to the customer” does not automatically create a sending tool. Review permissions and use the history to monitor results.

Your first exercise in FarolAI

Create an agent for one task, write short instructions and try three requests: one complete, one with missing information and one outside its scope. Refine the instructions until it handles all three well.

Then grant access to the right people and follow the WhatsApp setup guide to bring this workflow to your team’s conversations.

PUT IT INTO PRACTICE

Start with your first agent.

Give your team a new way to handle everyday tasks.

Create an account ↗