Skip to content

STRATEGICRepublic of San Marino

AboutContact
ItalianoEnglish
Client area
Contact

AI training courses · developers and IT teams

Advanced AI courses for developers and IT teams

The course is for people who write code and people who connect AI to company systems. You learn to use Claude, Claude Code and ChatGPT on your code. Then you use the APIs to build assistants and agents with clear permissions and proper defences. It runs remotely or on site and ends with a practical exercise.

Let’s talk about a course for your teamThe programme
Example of the practical exercise in the developers’ course. The code for an agent, written with the assistant and reviewed by a person, sets its permissions before work begins. In a test environment the assistant reads and cites its source, prepares drafts that a person confirms and cannot send anything to clients, and every action goes into the log. The test results also show an instruction hidden in a PDF that is not carried out.
Example · the code, permissions and activities shown are not real
Claude, Claude Code and ChatGPT
for coding and through the APIs
Assistants and agents
with permissions, logs and defences
Remote or on site
across Italy, in San Marino and at partner venues in Turin and Rome

01 · copy the link to this sectionThe situation

Connecting AI to company systems without opening security holes

Your team already uses AI to write code. Now it has to bring AI into your systems, and that is where the rules change.

Many developers already use Claude or ChatGPT to write and fix code, each in their own way. Then management asks for something more: an assistant that answers from internal documents, or an agent that reads orders and prepares them in the ERP.

At that point the questions are different. Which data can the assistant read, and under which account? What happens if a document contains hidden instructions? Who checks AI-written code before it goes live? What does each answer cost once requests run into the thousands?

The course answers these questions using your team’s real work, with exercises, tests and written rules.

Typical scenario

A developer connects an assistant to the document archive in a few days. It works, and management wants to open it up to everyone. Nobody has yet decided which folders it can read, who sees the answers, or what happens if a PDF contains instructions written to trick it.

The course programme

02 · copy the link to this sectionTwo uses

AI that helps you write code, and AI inside the software

The course keeps the two uses apart, because the risks are different. In the first, AI helps the person writing the code. In the second, it becomes part of the product.

AI

Claude, Claude Code and ChatGPT for writing code

You use the assistant to write, explain and review code. With Claude Code, the AI works inside the project. It reads the files, proposes changes and can run commands. You learn to give it small, clear tasks, to decide what it may do without asking, and to read every change before accepting it. Before it goes live, AI-written code is reviewed and tested like everything else. Not a developer? ChatGPT and Claude courses for office teams
AI

Applications built with the Anthropic and OpenAI APIs

With the APIs you build assistants that answer from company documents, agents that prepare operations in your systems and automations that read attachments. You learn to choose the model, write the standing instructions that guide the assistant, give it tools to use, and measure quality and cost. This is where permissions, logs and defences come in. The programme covers them one by one. The assistant for company documents: Company Brain

ChatGPT is a trademark of OpenAI. Claude and Claude Code are trademarks of Anthropic. We name them only to identify the tools the course covers.

03 · copy the link to this sectionThe programme

What you learn: six topics built around your team’s work

We start from what the team already uses and what it needs to build. The topics are chosen together, and not everyone needs all of them.

AI Assistants that cite documents (RAG) The assistant first finds the relevant passages in approved documents, then answers and shows where it found them. This technique is called RAG. You see how to prepare the documents, how to check that the answer really follows the source, and how to stop the assistant showing confidential files to people who are not allowed to see them.
AI Agents in your systems An agent does not stop at an answer. It reads a document, checks the data and prepares an operation in the ERP. You learn to give it written tasks and limited tools. Important operations are confirmed by a person.
AI Permissions and logs The assistant works under its own account, with the minimum permissions its task needs. It shows each person only what that person can already see. Requests and actions are kept in the logs, which help you understand an error and respond to an audit.
AI Defences against hidden instructions An email or a PDF can contain hidden instructions written for the assistant. This is called prompt injection, and it comes first on the list of the best-known risks of these systems, published by OWASP. OWASP is a non-profit foundation dedicated to software security. There is no single defence. You need several layers, and you learn to build them.
AI Measuring quality and cost You prepare a set of test questions with expected answers. This shows whether a change makes the assistant better or worse. The cost of each answer is estimated before the tool is opened up to everyone.
AI AI-written code, reviewed and tested Code proposed by the assistant goes through the same review and the same tests as code written by hand. You learn to ask for tests along with the code, and never to paste credentials or real data into tools that have not been approved.

The topics, and how much time each one gets, are agreed with the team lead before we start.

All AI training coursesFor database administrators: AI courses for DBAs

04 · copy the link to this sectionThe practical exercise

At the end, the team builds a small assistant and tries to trick it

The course ends with real work, not with a quiz.

The case is chosen together and can be a document assistant or an agent. The work takes place in a separate environment, with no access to live systems.

  • Test data. Fake or anonymised documents and data, modelled on the real ones.

  • Permissions decided up front. What it can read and what it can do, written down before we start.

  • Questions designed to trick it. We try to make it say what it shouldn’t and do what it can’t. Every hole we find becomes a fix.

  • Measured results. Test questions with expected answers, errors logged and an estimated cost.

The completed exercise stays with the organisation, together with the materials and the training record.

The training record

05 · copy the link to this sectionSecurity

Security checks are done by people who handle security every day

The course teaches you to design with care. Checks on live systems are a separate job.

Before an assistant or an agent reaches the company’s real data, the technical checks are carried out by Strategic’s Cybersecurity division. They are done with written authorisation and, where possible, in a test environment.

Development, databases and security work in the same group. If the project needs to be built after the course, Strategic Labs develops custom assistants, agents and integrations.

AI security in the Cybersecurity divisionAI agents and integrationsCustom software

06 · copy the link to this sectionWhere and how

Where the courses take place

  • Remote or on site. The course runs remotely, in live online sessions, or on site. On site, we come to you anywhere in Italy, or the sessions take place at Strategic’s premises in San Marino or at partner venues in Turin and Rome. The two formats can be combined, for example with the introduction held remotely and the practical exercise on site.

Locations and formats for all courses

07 · copy the link to this sectionFrequently asked questions

Frequently asked questions about AI courses for developers

Which is better for programming, Claude or ChatGPT?
We don’t rank them. The models change often, and what matters is how they perform on your tasks. In the course, the approved tools are tried on the same exercises against criteria written in advance: whether the code is correct, whether the tests pass, how long it takes to review and what it costs. If the tool has already been chosen, we work with that one.
Does the course cover Claude Code?
Yes, if your organisation uses it or has decided to adopt it. Claude Code is Anthropic’s coding tool. It works inside the project, reads the files, proposes changes and runs commands. You learn to decide what it may do without asking for confirmation, to give it small tasks that are easy to check, and to read every change before accepting it. The code it produces goes through the same review and tests as everything else.
Can we work on our own code?
Yes, if the organisation authorises it and the tool is approved for that code. First we agree with IT which projects to use and check what the contract with the tool’s vendor allows. If that is not possible, we work on a test project built on the same model, with fake data.
Do you use the Anthropic and OpenAI APIs?
Yes. We work with the API of the vendor your organisation has chosen, or with both if the team needs to compare them. You see how to write a request to the model, how to give it tools to use, and how to handle errors, usage limits and costs. Access keys are kept out of the code.
How do you stop an agent from causing damage?
The agent’s access is designed like that of an external supplier. It has written tasks, an account with the minimum permissions and no access to payments or bank details. Important operations are confirmed by a person, everything is logged, and a switch turns it off without losing anything. In the course, these rules are put to the test in the separate environment.
Do you also run courses for DBAs?
Yes. Database administrators have a separate course, in which AI has read-only access to the data. It covers SQL queries, execution plans, migrations and security, with examples on Oracle.

AI courses for DBAs AI agents and integrations Frequently asked questions about all our courses

08 · copy the link to this sectionObservatory

From the Observatory: assistants, agents and defences

Analysis and guides from the Strategic editorial team.

All articles on artificial intelligence

09 · copy the link to this sectionContact

A course for your technical team starts from your systems

A few pointers are enough: which languages and systems the team works with, which AI tools you already use and what you would like to build. We reply by email, usually within two working days.

Please do not put code, credentials or confidential data in the form.

Or write to us:WhatsAppstaff@strategic.sm

On WhatsApp, messages are answered by Strategic’s AI assistant. Please do not use it for confidential information.

Your requestAIAI training courses

Do any of these questions sound familiar? tick the ones that apply to you

The form opens with the topic and the questions already filled in.