Strategy / AI-ready software
The agent is the last piece.

There is a lot of talk about AI agents that automate everything. In a real business, an agent is only as reliable as the processes and tools it works with. That is why the order is reversed: first the process, then a custom tool, then instructions for agents. Agents come last, if they are needed, and people stay in charge.
- Starting point
- The real process, with its people and exceptions
- What is built
- A custom tool with APIs, an MCP server and skills
- Towards
- Useful agents, where needed, governed by people
01 / The promise
The problem is not the agent.
“Add an agent and automate everything” works in demos. In a business, the agent meets unclear processes and programs designed for people, not for agents: it has to guess, and it gets things wrong.
The point is not choosing the best agent. It is preparing the ground it works on.

02 / The strategy
Four steps, in this order.
The order matters more than the technology. Each step supports the next.
- 01
Understand the process.
How the work really happens: steps, people, tools and exceptions. Often the first improvement is reorganising, with no new software.
- 02
A small, custom tool.
Where needed, a tool is developed with one clear purpose, for that business. AI is already at work here: it speeds up development and makes custom software affordable for an SME.
- 03
Instructions for agents.
Every tool comes with documented APIs, an MCP server and dedicated skills. In practice, a user manual written for agents.
- 04
Agents, if needed.
The tool is useful without them. When the business decides to hand some steps to an agent, nothing has to be rebuilt.
03 / AI-ready
A user manual written for agents.
A compatible agent does not have to guess: it knows what it can do, in what order and with which data, the same way every time.
- 01
Documented APIs
The doors through which the tool’s data is read and written. Every operation has a name, parameters and clear permissions.
- 02
MCP server
MCP (Model Context Protocol) is an open standard for connecting AI assistants to tools. With an MCP server, a compatible agent can see which operations it can carry out and how. The business is not tied to a single AI provider.
- 03
Dedicated skills
Instructions written for agents on how to use the tool in a specific case: in what order, with which checks and what to do when something does not add up.

04 / Control
People stay in charge.
- The agent can only do what the tool allows, with dedicated, limited access.
- Actions that matter, such as changing data or sending something outside the business, are approved by a person.
- Every operation is logged: who requested it, what was done and when.
- Agents can be switched off at any time: the tool keeps working.
Summary
Two ways to bring AI into a business.
| Hype | Strategy |
|---|---|
| Start from the agent | Start from the process |
| Generic software | A custom tool |
| The agent guesses | The agent has instructions |
| “Fully automatic” | People decide |
05 / Questions
Frequently asked questions.
What is an AI agent?
An AI assistant that does more than answer: it uses tools to carry out steps, such as looking up data, preparing a draft or updating a record.
Do we have to use agents?
No. The tool is designed first of all for the people who use it every day. APIs, the MCP server and skills are built together with the tool, so the option stays open without rebuilding it.
Which AI assistants does it work with?
Those compatible with MCP, the open standard adopted by the main AI providers. The business can choose and change provider without touching the tool.
What happens to company data?
Data stays in the tool, and the agent only sees what its permissions allow. Which AI services to use, and where data travels, is assessed beforehand, case by case.
Which process would you start with?
A first conversation. For an initial phone call, all you need is the process you would like to improve and the tools you use today. There is no need to have decided whether to use agents.
Tell me what you would like to improve. We’ll work out together where to start and whether I can help.
