Available · 2 projects · Q4 2026Independent practice · Montreal

AI agents · Quebec SMBs

An AI agent that does a real task — and asks before it acts.

In practical terms, an AI agent is a software assistant that receives a request, pulls the information from your systems, prepares the action — a reply, a quote, a data entry — and submits it to you. Not a chat gadget: a virtual employee assigned to one specific task, working with your data and under your rules.

Gabriel Nadon, independent consultant in Montreal. You talk to the person who builds it.

Agents that pay off in an SMB
  • Intake: reads every request that comes in by email, creates the record and drafts the reply.
  • Quotes: builds a first draft from your prices, templates and past projects.
  • Customer service: answers repetitive questions from your documents, and hands off to a person.
  • Follow-up: spots overdue files and prepares the reminders.

Why an agent, and not just ChatGPT

ChatGPT helps one person. An agent takes a task off the company’s plate.

A ChatGPT subscription makes each employee a bit faster — as long as they remember to use it. An agent is wired into the process itself: it starts on its own when a request comes in, and it knows your data.

  • The same customer questions come back every day, and someone answers them by hand every time.
  • Requests arrive by email, web form and phone — and get lost between the three.
  • A quote takes hours even though most of its content already exists somewhere.
  • The information for a file is scattered across emails, PDFs and software.
  • You tried a generic chatbot: it made up answers.

How I work

A good agent has one task, access to the right data, and a clear limit on what it can do on its own.

The agents that fail are the ones asked to “do everything.” I design narrow agents: one task, defined information sources, written rules, and a human approval step before any action that commits the business — sending, paying, making a commitment to a customer.

  • It answers from your data

    The agent cites your documents, your prices, your files. If it can’t find the answer, it says so and hands off — it doesn’t make things up.

  • It asks before it acts

    Draft, proposal, pending entry: the final decision stays with a person until trust is established.

  • It leaves a trail

    Every action is logged: what it read, what it proposed, who approved it.

How it works, in five steps

  1. Pick the task

    A frequent, repetitive and costly task. We put a number on it before writing a single line.

  2. Gather the sources

    Documents, templates, prices, history: what the agent needs to know, and what it must never see.

  3. Prototype on real cases

    We replay past requests. You compare what the agent proposes with what your team actually did.

  4. Supervised go-live

    The agent prepares, your team approves. We measure the time saved and the corrections needed.

  5. Expand carefully

    Once its proposals are reliable, we give it more autonomy — or a second task.

Real results

I run my own agents before selling you one: my prospecting system finds companies, analyses their website and drafts a personalised first email — nothing goes out without my approval. On the client side, the supplier price system of an independent grocery store reads every supplier’s price list and prepares the point-of-sale update, which the team approves: close to $56,000 a year of data entry eliminated. Read the full case study →

Pricing

First agent: fixed-scope sprint from $4,500, 2 to 3 weeks. On top of that come the AI model’s usage fees, billed per use by the provider — I estimate them on your real volumes before building, so there are no surprises.

Questions I often get

What’s the difference between an AI agent and a chatbot?

A chatbot converses. An agent acts: it checks your systems, prepares an entry, a reply or a document, and fits into the process. A customer service chatbot can in fact be an agent, if it answers from your documents and hands off to the right person.

Can an AI agent make mistakes?

Yes, like a new employee. That’s why it answers from your sources, flags when it isn’t sure, and a person approves any action that commits the business.

Isn’t ChatGPT Enterprise or Copilot enough?

To help each employee write or summarise, often yes. For a task to get done without anyone thinking about it — as soon as an email arrives, for example — AI has to be connected to your systems and your rules. That’s what an agent does.

What happens to our data?

We define precisely what the agent can read and write. Personal information is handled according to your obligations (Quebec’s Law 25): providers chosen accordingly, limited access, and none of your data used to train a public model.

How long before an agent is useful?

A first narrow agent goes live under supervision in 2 to 3 weeks. The value shows up as soon as it correctly prepares most cases, even if your team still approves every proposal.

Free review

What task would you hand to one more employee?

Describe it. Within 24 hours I’ll tell you whether an AI agent can take it on, with which data and how much effort — whether or not we end up working together.

Or book your 20 minutes now

Reply within 24 hours. Or book 20 minutes directly.