Create your first AI agent in ten minutes, without installing anything
By the end of this tutorial, you will know how to create an AI agent that handles a specific task from your week. You will know what to tell it, which documents to attach, how to read its plan, and how to test it safely. No lines of code, no installation: your words and your files are enough.
- Start with a single repetitive, well-defined task.
- Describe it in your own words and attach the documents you already use.
- A good agent asks you questions before working: answer them precisely.
- The dry run shows the actual work without sending anything.
- You validate the first sends manually, and you always keep the final say.
Creating an AI agent no longer requires a developer. What matters most is knowing what you expect from it. If you don't yet know what an agent is, or how it differs from a chatbot, first read our two definition articles. Here, we move to practice: the complete journey, step by step, with each time what the agent does and what you decide. Plan for about ten minutes of your time, a bit more if your documents are scattered.
IThe journey at a glance
On ABYS, creation always follows the same order. This order is not a detail: it prevents the agent from working blind and prevents you from discovering a problem the day a client receives an email.
- 1Say what you want
You describe the task in your own words, as you would to a new employee. No jargon, no form to fill out box by box. One clear sentence is worth more than a long fuzzy text.
- 2Attach your documents
You add the files you already use to do this task: PDFs, spreadsheets, photos. The agent relies on them rather than on assumptions.
- 3Answer its questions
The agent identifies what it is missing and asks you for it before working: a deadline, a tone, a threshold, a special case. Your answers complete the mission.
- 4Read its plan
The agent proposes a plan: what it will read, what it will produce, when, and what it won't do. You correct it until it suits you.
- 5Test dry run
The agent makes a real run on your real data, but nothing is sent. You see exactly what it would have done.
- 6Choose the schedule and validate
Your plan sets the frequency of work for your agents, then you validate their first sends yourself before they go out.
IIChoose the right first task
The first agent succeeds when the task is repetitive, bounded and verifiable. Repetitive: it comes back each week. Bounded: you know where it starts and where it stops. Verifiable: by reading the result, you can immediately see if it is correct. Follow up on overdue invoices, prepare a quote from a grid, write the week's menu: these are good candidates. "Manage my communication" or "help me with my business" are not, because no one could say when the work is finished.
IIIStep 1: describe the task in your own words
Write as you would speak to someone you trust who is arriving Monday. Say what needs to be done, for whom, based on what, and what you absolutely don't want. The prohibitions matter as much as the task: "never offer a discount", "don't write to customers on the weekend". No need to polish the style. The agent will rephrase and, if it's missing something, it will tell you in the next step. To learn more on this point, our article on writing the task details a five-point method.
IVStep 2: attach the right documents
An agent without documents invents or stays vague. An agent with your documents works like you do. Attach what you already consult when you do the task by hand. A PDF, a spreadsheet or even a photo of a clearly legible sheet will do. Only attach what serves the task: a pile of unrelated files confuses the work and consumes tokens for nothing.
Check what applies to you: the list remembers your answers on this device.
VSteps 3 and 4: the agent's questions, then its plan
An ABYS agent asks for everything it needs before working. This is intentional. For the carpenter, the questions might be: "Does your grid include installation?", "What is the validity period for the quote?", "What should I do if the request doesn't provide dimensions?". Answer briefly and precisely. Each answer prevents a mistake later on. If a question surprises you, it's often because it touches on a case you handle in your head without ever having written it down.
Next comes the plan. Read it like a contract: what the agent will read, what it will produce, when, and where it will stop. Check three things. Is the scope right? Are the prohibitions in it? Are the cases where it should hand over to you written down? If something is off, say it simply, and the agent corrects the plan.
VIStep 5: the dry run, your safety net
Thedry run is a complete run on your real data, during which nothing is sent. You see the emails it would have written, the quotes it would have prepared, the cases it would have ruled out. This is the time to read line by line. A rate applied incorrectly, a tone too casual, a customer it should not have contacted: everything is corrected here, without consequence. If you are not satisfied with the result, adjust the mission or a document, then run another test.
VIIStep 6: set the schedule and keep the final say
On ABYS, your plan sets the pace: once a week at 19 euros excl. VAT per month, every day at 99 euros, every four hours at 199 euros, every hour at 349 euros. An email received by the agent makes it process without waiting for its next turn. A tighter pace is not necessarily better: each processing consumes tokens, drawn from the monthly pool shared by all your agents. For follow-ups, one pass per week is often sufficient.
At the start, you validate the first sends manually. After three validations, you can switch the agent to automatic validation, and the risks of this choice are then spelled out in black and white. Nothing requires you to do so. For a task that affects your customers, I advise you to keep control for a few more weeks, to see rare cases come through.
A carpenter is hesitant between several tasks for their very first agent. Which one has the best chance of succeeding?
?Questions we're asked
Can you create an AI agent for free?
You can test instructions for free in an assistant like ChatGPT or Le Chat, but it will not work alone at a fixed pace. On ABYS, an agent that works for you requires a subscription, from 19 euros excl. VAT per month, with multiple agents possible. The free AI audit can help you identify the right first task.
Do you need to know how to code to create an AI agent?
No. You describe the task in English, you attach your documents and you answer the agent's questions. What matters is the clarity of the mission and the quality of the documents, not the technique. A craftsperson who knows how to explain their trade to an apprentice already has the main skill.
How long does it take to create an AI agent?
For a simple task and ready documents, allow about ten minutes for the description, questions and plan. Add the time to carefully read the blank test. If your rates or conditions are not written anywhere, gather them first: this is often the longest part.
What happens if the agent makes a mistake?
During the dry run, nothing is sent: you correct the task or the document in question. Then, as long as you validate manually, no error goes out without your approval. This is precisely why it's better to keep manual validation at the start.
Do you have a task in mind that comes back every week? Describe it in your own words and see the plan the agent proposes to you.
Create My AgentRead more next
What documents to give an AI agent, and which to keep
An AI agent works from what you give it. This guide tells you which documents really help it, in what form…
Read the lessonWriting an AI agent's mission: the five-point method
An agent works as well as their mission is clear. This course gives you a five-point method to write it: the goal, the…
Read the lessonHave the final say on an AI agent: validate, then delegate
An AI agent prepares the work, you decide what goes out. You'll know at the end how to review their first sends, what signs…
Read the lesson