AI agents for in-house IT in SMEs

Today

The same questions come back every week: passwords, printers, remote access, sharing a Teams folder. They keep interrupting the person who runs IT.

With AI

We configure an agent in Teams connected to your knowledge base, your ticketing tool and, read-only, to Intune. For each request it checks the status of the device concerned, answers when an approved procedure exists, and otherwise opens a triaged ticket (category, priority, device, steps already tried) with a proposed fix. Anything involving access or security goes to the person responsible for IT, who also reviews the standard answers; every exchange is logged in the ticket.

Today

Documentation of the network, servers and procedures is incomplete or out of date; it lives mostly in the head of the administrator or the provider.

With AI

The administrator places configuration exports with secrets removed, along with notes and scripts, in a repository. A ready-made agent such as Claude Code or GitHub Copilot in agent mode works through them, writes structured documentation and step-by-step procedures, flags gaps against the existing documentation and opens a pull request. The administrator checks each procedure by running through it before merging and publishing.

Today

The development team spends a lot of time on repetitive code, writing tests and making sense of old, poorly documented code.

With AI

We roll out coding agents (Claude Code, GitHub Copilot in agent mode, OpenAI Codex), excluding sensitive repositories, and train the team to hand them a task described in a ticket: the agent reads the code, writes the change and the tests, runs the test suite and opens a pull request explaining what it did. Every pull request goes through human review and continuous integration before merging; the technical lead decides on release to production.

Today

Microsoft patches, business application updates and vulnerability advisories pile up. Working out which devices and servers are affected, and in what order to act, takes hours of research.

With AI

Whenever patches or vulnerability advisories are published, the agent cross-checks the Intune and Defender inventory against the release notes, lists the devices and servers affected, flags known incompatibilities with your applications and prepares a phased rollout plan as a change request in the ticketing tool. The administrator approves the plan and starts the rollout personally; the agent then tracks device status and reports those still lagging behind.

Today

With every new hire, role change or departure, permissions are adjusted case by case. Unused accounts and overly broad access linger.

With AI

When HR announces a new hire, role change or departure, the agent reads the role profile, prepares the matching groups, licences and application access in Entra ID as a request awaiting approval, and creates the ticket to set up the workstation. For periodic reviews, it compares actual permissions with the profile and lists discrepancies. The administrator approves every grant or removal, the relevant manager signs off the review, and the agent logs each of its actions.

Today

Staff forward dubious e-mails, and security tools produce alerts that nobody has time to read in full.

With AI

For each reported e-mail or Microsoft Defender alert, the agent investigates according to your procedures: it analyses headers and links, looks for the same message in other mailboxes, checks recent sign-ins for the account concerned and writes a summary in the ticket with severity, evidence and recommended measures. The person responsible for security decides on the measures (deletion, blocking, resetting, informing staff); the agent does not act on accounts by itself and logs every lookup.

Today

Sales, stock and footfall figures are split between the till, the online shop and the ERP. Every question from management needs a manual Excel export.

With AI

We build a database bringing these sources together and an agent that, on a question asked in Teams or every Monday, queries it through controlled queries, reconciles till, online shop and ERP, flags discrepancies, produces the report with its charts, files it in SharePoint and posts it in the management channel. The person in charge of data approves the definition of each indicator, and every figure cites its source and query.

Today

Staff want to query files with AI, but each client file must remain visible only to authorised people, as professional secrecy requires.

With AI

We build a search assistant that respects, file by file, the access rights of your file server or SharePoint and logs every query. IT validates the permission setup through testing before launch, and the partners decide which files go in.