CrafterAgency
Introductory callBook

AI agents that actually work. Not demos

Document handling, replies to requests, research, quote preparation: I build AI agents that do one precise task, connected to your tools, with your approval where it matters.

Quote agentRequest no. 2026-318
Email from Mr Garnier: bathroom renovationDone
Client found in the CRM, 2 previous jobsDone
Measurements read from the attached planDone
Quote prepared: 6 lines, 8 420 excl. VATDone
Send the quote to the clientTo approve
Approve sendingEdit
An agent does one precise task, and you keep control of what gets sent. Illustrative example.

What an AI agent actually is

An agent is a program that uses an AI model to act inside a defined process: it reads what it receives, chooses the next step, uses your tools, and returns a structured result.

An agent works when it has one clearly defined job. It fails when you ask it to do everything. I only build single-task agents, and I measure their results before I put them into service.

  • Sort a document
  • Summarize a meeting
  • Prepare a quote
  • Route a client request

The agents I build

Always one precise task, always connected to your tools.

Document handling

Invoices, contracts, forms, tenders: extract, sort, forward, and flag the exceptions.

Research and writing

Summaries, competitor monitoring, and first drafts with their sources.

Client relations

Sorting requests, qualifying prospects, and drafting replies, with a handoff to a person when needed.

Internal tasks

Meeting reports into the CRM, calendar coordination, and knowledge base updates.

Multi-step processes

Agents that chain emails, databases, software, and documents, from the start of a case to the end.

How I work

The same pace as the Growth plan: an audit, then a delivery each month.

  1. Step 1

    Define the task

    You and I set in writing what the agent receives, what it must produce, and which tools it uses.

  2. Step 2

    Build a solid base

    If it can be done without AI, that is how it is done. AI only steps in where it adds something.

  3. Step 3

    Test on your real cases

    A set of real cases is the test for every change, to measure quality and spot drift.

  4. Step 4

    Put into service and monitor

    Monitoring, alerts, and a human check where you need one, in line with the European AI Act.

What it looks like in practice

Three common situations, so you can picture it.

Trades and construction

Quote preparation agent

It reads the request and the attached plans, finds the client, and prepares the quote. You approve it before it is sent.

Real estate and management

Lease analysis agent

It extracts the key terms, amounts, and risks from each lease. Unusual cases are reviewed by a person.

Business services

Meeting follow-up agent

It turns the notes into a report, creates the tasks, and updates the client record.

Illustrative examples of common situations, not client references.

Two ways to start, at a fixed price

Prices shown for
A first project10 500 € excl. VAT

A single-task agent, in production: defining the task, building it, tests on your real cases, monitoring, and adjustments after launch.

Fixed price, with no extras. You move to the subscription only if the result suits you.

Start with a project
Growth plan4 700 € excl. VATper month

The opening audit, then a delivery each month on this topic or on all your other tools, plus maintenance and training for your teams.

Launch offer: 3 200 € excl. VAT per month
For the first 4 clients, for the first year, the first month is free. 3 places left.
No minimum term, 2 months' notice.

Also budget for this, paid directly to the vendors: AI model and hosting fees, about 50 € to 500 € per month depending on use. I price them exactly during the audit.

Common questions

What is the difference between an agent and a chatbot?

A chatbot answers questions. An agent acts: it uses your software, reads your databases, and writes into your tools.

Can the agent be wrong?

Yes. I build it so mistakes are rare, minor, and caught: checks, cited sources, regular tests, and a human check for important decisions.

Which AI models do you use?

I choose based on the task: Claude for reasoning and documents, GPT, or open-source models such as Mistral or Llama when hosting or cost requires it.

Can it run on our own servers?

Yes, with open-source models on your servers. For an online service, I use services hosted in the European Union when they exist.

Will it need retraining?

Rarely. Improving the instructions and the tests usually works better than retraining.

Combine with

Assistants on your documentsWhen you need to answer questions rather than act.See the pageProcess automationWhen the agent is only one step in a wider flow.See the page

Do you have a task built for an agent?

A free introductory call, to see whether I can save you time. You leave with specific next steps, whether you work with me or not.

Book an introductory callCall [00 00 00 00 00]

Free and no commitment. The button opens your email with a message ready to send. You can also write directly tocontact@votre-domaine.fr. Reply within [24 business hours].

Your message already includes three questions to prepare the call

  1. How many people work in your company?
  2. What sector are you in?
  3. Which task wastes the most time for your team?