AI & Agentic Systems

AI is no longer a promise, it's already inside companies

AI and Agentic Systems consulting for businesses

We don't sell tools, we don't sell training. We build with you a bespoke transformation path, where AI enters your processes with clear objectives and verifiable results. We partner with the company — integrating technology, organisation and governance — not as a vendor of a tool that ends up gathering dust on a desktop.

— Defining the terms

An agent is not a chatbot · it's a process

An agentic system is software that observes a state, decides what to do, acts, and verifies the outcome — autonomously, within boundaries you define. Unlike a conversational assistant, an agent doesn't wait for someone to ask it a question: it works like a colleague executing a process, within the rules you've given it.

When it makes sense, agentic AI orchestrates the interaction across ERP, CRM, portals, document systems and data platforms — it doesn't live in a silo, it works inside the infrastructure that already exists, reducing redundancies and increasing information coherence.

This changes the way of thinking about AI in business. No longer "we use it to ask things", but "we embed it in an operational flow, give it access to the data it needs, put security and compliance guardrails around it, and measure whether it's producing the value we expected".

— 01 / Integrated practices

Four practices, a single integration

When a company introduces AI, it usually buys just one dimension — the model, the automation, the chatbot. The real value arrives when the four dimensions are governed together: processes, security, privacy, development. That's how we work.

Processes

Agentic Workflow

We identify where AI really shifts the work

We start from your process, not from an abstract use case. We map the end-to-end flows, identify the points where an AI agent would reduce time, errors or costs without creating risky dependencies. For each candidate we define the decision boundary, the data it will have access to, the humans who remain in the loop, and the criteria for declaring the experiment a success or a failure.

Process mining and candidate identification
Agentic workflow with explicit human-in-the-loop
ROI measurement before and after, on a shared baseline

Security

Information Security

An agent that accesses your data is a privileged endpoint

When an agentic system interacts with your information systems, it inherits new risk surfaces: prompt injection, data leakage through responses, permission escalation via tool use, exfiltration via chain-of-thought. We work on the security posture with the rigour we apply to other critical systems.

Risk assessment for integrated AI systems
Threat modelling for agentic architectures
Hardening of tools and MCP servers
Alignment to NIS2 and ISO 27001 with AI extensions

Privacy

Data & GDPR

Your company's data or your clients' data that ends up in the prompts

The real problem isn't theoretical. An AI agent running on your processes sees strategic data, client data, confidential information, and in many cases sensitive data or data protected by professional secrecy. All of this enters the prompts, passes through models often hosted outside the EU, ends up in vendor logs, and can persist in the agents' persistent memories. Not GDPR-by-the-book: with the discipline needed to defend the position before a regulator, an angry client, or a supervisory body.

Mapping of the data each agent can see
Cross-border flows and models hosted outside the EU
Retention and persistent memory of agents
Professional secrecy, health data, minors' data

Development

Code & Production Readiness

From vibe coding to code that ships to production

Your developers already use AI agents to write code. It's good news for productivity, but AI-generated code has new properties that your release process wasn't designed to absorb. We don't tell you to slow your developers down. We help you verify that the development model in use actually ships to production code that will hold up to an audit or an incident.

Review of development processes that use AI assistants
Traceability of what is written by humans and what by AI
Security of generated code (vulnerabilities, secrets, supply chain)
IP and licensing of code produced with AI

— 02 / Method

How we bring AI inside a company

Discovery & framing

1–2 weeks

Conversations with decision-makers and operational teams, a reading of your information architecture, mapping of the regulatory landscape specific to your sector. The output is a 15–20 page document: three candidate processes, three risks you hadn't yet brought into focus, a recommendation on where to begin.

Disciplined pilot

6–8 weeks

On one process only. We define the scope, we build the agent with security and privacy guardrails from the first prototype, we run it in parallel with the human process to measure the difference. At the midpoint of the pilot there's a formal check: we proceed, we recalibrate the scope, or we close the hypothesis. Without forcing it.

Industrialisation & governance

Ongoing

When the pilot works, we take it to production: stable integrations, production monitoring, operational processes for the humans working alongside the agent, governance framework for the decisions that will come after this one. We stay as advisors as long as needed; we leave when your team is autonomous.

— 03 / Positioning

Three things we don't do

N° 01

We don't resell AI models

No exclusive partnership with OpenAI, Anthropic, Google or others. We choose the right tool for your problem, and if the landscape changes we change the tool.

N° 02

We don't promise "AI transformation"

We tell you what AI can do in your specific process, within what timeframe, with what margin of risk. If we don't see any, we tell you.

N° 03

We don't sell POCs that stay on the desktop

Every pilot we run is designed to go to production if it works, or to be cleanly closed if it doesn't.

— Want to understand where AI can actually help you?

Let's talk — One hour of conversation, zero vendor slides

Tell us about your context: sector, data you handle, processes that weigh on you. We'll give you an honest read on where an AI agent would make a difference, where instead it would be costly noise, and which security and privacy pre-conditions need to be sorted out before even beginning.

Let's talk