AI agents
AI agents for customer service, lead qualification, document synthesis for SMBs — evaluated on what they replace, not on what they promise.
Concrete impact
Available 24/7
An answer, even while you sleep
Customers get an instant reply, at any hour. No weekends, no holidays, no queues.
Consistent quality
Same tone, same level, every interaction
No bad days, no turnover, no variation between two employees. The experience is the same for everyone.
Instant response
From several hours to a few seconds
Repetitive requests are handled before they reach your desk. Your teams focus on what needs a human.
Scoped & measurable
Evaluated on what it replaces
Not a gadget. Every agent ships with an evaluation framework, guardrails and monitoring. You know what it does well — and what it doesn't.
This solution is for you if…
- //You handle a volume of repetitive requests (support, qualification, FAQ) that overwhelms your teams.
- //You have documents, data, customer feedback that nobody has time to read in detail.
- //You want to test AI on a concrete case, with a scoped perimeter, before investing more broadly.
- //You're looking for a partner who understands what AI does well — and especially what it does badly.
What it costs you today
The AI landscape has become a circus: everyone's selling "the agent that does everything", and most demos don't survive the first real use case. Result: many companies had their first AI experience with a poorly scoped tool that disappointed — and quite reasonably cooled off.
An agent that works is the opposite: a narrow scope, concrete evaluations, a gradual production rollout. You don't replace a team with an agent; you take the tedious part of the work off its plate. The agent that answers 70% of repetitive questions in 10 seconds is the one that has value — not the one that claims to do everything.
A few concrete examples
A level-1 support agent that answers your customers' recurring questions, connected to your knowledge base.
A lead qualifier that analyzes incoming requests and scores them against your criteria.
An internal assistant that searches through your documents, contracts or meeting notes.
An agent that reads and summarizes customer feedback (emails, reviews, tickets) and surfaces weak signals.
An operational copilot for a specific role: sales, HR, legal, according to your need.
A sales writing assistant that drafts quotes and proposals from a short brief.
An incoming-mail triage and summary agent that prepares your replies and prioritizes urgency.
My 4-step method
Understand
Thirty minutes to understand your context. I tell you straight whether I'm the right person.
Scope
Within 48h, a fixed-price quote, precise scope, deadlines kept. No fuzzy hourly rate.
Build
I develop, I show you intermediate versions, you adjust as we go.
Hand over
I deliver the code, the documentation, a handover session. What I build belongs to you.
from €1,500
Fixed price, no surprise overruns. Precise quote within 48h.
What people ask me most
How long to get a first agent in production?+
A targeted agent can run in production in 4 to 8 weeks, with an honest evaluation phase. I don't accept to deliver an agent whose quality I can't measure.
How much does it cost to run?+
An AI agent makes calls to the models (Claude, GPT, Gemini, Mistral…), billed per use by the provider. The cost depends on the volume of requests — I give you a monthly ballpark during scoping, and we pick the model with that in mind too. You pay those costs directly to the provider; I never mark up on them.
Which LLMs do you use?+
I choose based on the need: Claude, GPT, Gemini, Mistral, open-source models. The right model depends on the case — accuracy, cost, latency, sovereignty. I explain the choice and leave you the option to pivot if the landscape shifts.
What if I want to switch models later?+
It's accounted for. The AI landscape moves fast — a better, cheaper or more sovereign model can appear at any time. The agent is built so we can pivot without rebuilding everything, and I explain the trade-offs of each choice.
Can the agent connect to my tools and knowledge base?+
Yes — that's the whole point of a custom agent: it draws on your documents, your CRM, your ticketing, your processes — not generic answers. We define together which sources it can access, and which it can't.
What if the agent hallucinates or gets it wrong?+
It's planned from the design stage: guardrails, reminders, human-in-the-loop on sensitive decisions, response monitoring. An agent that works is an agent whose good and bad outputs you can always see.
What about your confidential data?+
We define that framework from the start: hosting, retention, access, choice of provider. Depending on the needs we can keep everything in Europe, or use models that retain no data. No magic — only explicit decisions.
Who owns the agent and its code?+
You do, from delivery — code, configuration, prompts, documentation. You can evolve it yourself, ask me again, or hand it to someone else. No lock-in.
Another question? Ask it directly — I reply within 24 business hours.
A need like this one? Let's talk.
Thirty minutes to check it's relevant. No commitment, honest, and useful even if we don't end up working together.