Many small business owners have tried a conversational assistant, got a few useful answers, then gone back to their old habits. It is not a lack of motivation: a one-off conversation simply changes nothing about how the business runs day to day. What changes things is a task that gets done on its own, every day or every week, without you having to think about it.
This plan aims precisely at that: in 30 days, going from zero to one or two agents that actually run, that you have checked, and whose usefulness you know how to measure. It takes about an hour a week, a little more in the second week.
Before you start: what you need#
- A notebook or shared document to record tasks and observations.
- Access to your main tools: email, calendar, invoicing software, CRM or customer spreadsheet.
- A point person if you are not on your own: someone who will monitor the agents with you.
You do not need a big budget, a contractor or technical skills. A simple agent on Future Agent starts at a few euros a month, and the Free plan costs nothing beyond your active agents.
Week 1: spot the right tasks#
Days 1 to 5: the repetitive task log#
For five working days, note every task that comes back and annoys you a little. For each one, write down:
- what you do (in one sentence);
- how often;
- roughly how long it takes;
- the tools you use;
- what happens if you forget it.
Do not aim to be exhaustive. Fifteen to twenty lines are enough.
Days 6 and 7: the sorting grid#
Then sort the tasks using this simple grid:
| Criterion | Question to ask | Good candidate if… |
|---|---|---|
| Rule | Can I explain the task in two sentences? | Yes |
| Frequency | Does it come back at least every week? | Yes |
| Risk | Would a mistake have serious consequences? | No |
| Data | Is the information in a digital tool? | Yes |
| Visibility | Does the result go straight to a customer? | Preferably not, to start with |
Tasks that tick all five boxes are your first candidates. Typically you find things like: the press or market review, sorting emails, filing incoming invoices, the weekly sales recap, meeting minutes.
Week 2: launch the first agent#
Choose a single task#
Take the simplest candidate, not the most profitable one. The goal this week is to learn to describe, check and adjust. Once that reflex is in place, the next agents will go much faster.
Write the brief#
A good brief fits in a few lines and answers four questions: what, when, with which tools, in what form. Here are three examples ready to adapt.
Every morning at 8 am, read the emails received since yesterday in my Gmail inbox, sort them into "urgent", "to handle this week" and "for information", and send me a summary of ten lines maximum.
Whenever I receive a supplier invoice by email, save the PDF in the Google Drive folder "Invoices 2026 / current month" and add a line to the "Invoice tracking" spreadsheet with the supplier, the amount including VAT and the due date.
Every Monday at 7 am, put together a press review on energy-efficient renovation in France: five articles maximum, published the previous week, each with a two-sentence summary and the link. Send it by email.
These three briefs match existing templates in the library: inbox sorting, filing incoming invoices and morning press review. Starting from a template keeps you from forgetting a detail.
Let the architect ask its questions#
On Future Agent, you paste this brief into the chat. The AI architect asks the missing questions (delivery address, destination folder, exceptions), asks you to connect the necessary tools, shows you a sample result and tells you the monthly price. Nothing runs until you have approved it.
The first three days of operation#
Review every result. Note what is wrong: a misfiled email, a summary that is too long, an off-topic article. Adjust the agent's description with explicit rules, for example:
- "Emails from my accountant are always urgent."
- "Do not include articles behind a paywall."
- "The summary never exceeds ten lines."
Week 3: check, then launch the second agent#
The checkpoint#
After a week of operation, ask yourself three questions:
- Is the result usable without edits at least four times out of five?
- Have I stopped doing the task by hand? If you keep doing it "just in case", the trust is not there yet: work out why.
- Have I noticed a serious error? If so, add a rule or a human approval step.
If the answers are satisfactory, move on. If not, take one more week: one reliable agent is better than two approximate ones.
The second agent: one step further#
Now choose a slightly more demanding task, for example one whose result goes to a customer, but with your approval. Examples:
- an agent that prepares replies to frequent emails as drafts you review before sending;
- an agent that prepares quote follow-ups and submits them to you every morning;
- an agent that writes Google review replies that you publish after reading them.
The principle: the agent does the work, you have the final say. Over time, you can decide to drop the approval step for some simple cases.
Week 4: measure and decide what comes next#
Take stock with figures#
Go back to your week 1 log. For each automated task, compare:
| Task | Estimated time before (per week) | Time spent now (checking) | Oversights or delays before | Oversights or delays now |
|---|---|---|---|---|
| Email sorting | ||||
| Quote follow-ups |
Fill it in with your own observations. This table does not need to be accurate to the minute: it should let you decide honestly whether the agent is worth its price.
Choose the next three tasks#
From your initial list, select the next three tasks. You can now aim for more ambitious topics: a monthly dashboard, a cash position update every Monday, or a chain of agents handing work over to one another.
Share with the team#
Spend ten minutes presenting what has been set up and what it has changed, and ask everyone whether they have a task to suggest. The best ideas often come from the people who do the work every day.
A concrete example: the plan followed by a self-employed electrician#
To make the plan more tangible, here is how it might unfold for an electrician who works alone with a part-time assistant.
- Week 1: his log shows about fifteen tasks. The most frequent: answering repair requests arriving by email and through the website form, following up on quotes, sending the invoice after each job, reminding customers of their annual servicing.
- Week 2: he starts with the least risky task, sorting call-out requests. The agent reads incoming messages, sorts them by urgency and by town, and sends a recap every evening at 6 pm. In the first three days, he corrects two rules: "a water leak near an electrical panel is always urgent" and "requests from outside my area go in a separate section".
- Week 3: he adds an agent that prepares the invoice after each job from his voice notes. The assistant reviews it before it goes out.
- Week 4: the review shows that requests no longer get lost in the inbox and that invoices go out the same day. Next, he picks annual service reminders, then quote follow-ups.
Nothing spectacular, but two daily irritations have gone and the method is in place.
The classic pitfalls of these 30 days#
- Starting with the most complicated task because it is the most painful. Its turn will come, but later.
- Not reviewing the first results. An agent that makes mistakes silently for a month does more damage than an agent that does not exist.
- Writing a brief that is too vague, such as "handle my emails". The more precise the instruction, the better the result.
- Launching everything at once. You will no longer know which agent is causing the problem.
The article AI automation mistakes to avoid details these pitfalls and others.
If you would rather use WhatsApp#
If you are often out in the field, you can describe your task directly on WhatsApp at +33 7 56 81 76 53. It works the same way as on the website: the architect asks its questions, sends you a link to connect your tools and shows you the price before launching anything.
In short#
- Week 1: log your repetitive tasks and sort them with the grid.
- Week 2: launch a single simple agent, review everything, adjust.
- Week 3: check reliability, launch a second agent with human approval.
- Week 4: measure, choose what comes next, involve the team.
Thirty days will not transform your business. They will, however, give you a method, two agents working for you and a clear idea of what you can hand over next. To go further, the templates in the library can be filtered by sector and by profession, and use case pages, such as the one for tradespeople, show examples suited to a specific business.