Home/How we work with Claude
How we use Claudeand what we check
We work with Claude, the AI assistant developed by Anthropic, on almost all our projects: web applications, integrations, server scripts, firmware and documentation. We do not use it to skip work, but to work more carefully: every step is written down, tested and reversible.
Tool and component
On this page we explain in concrete terms what Claude does, what we do and where the line is.
The tool we build with
Claude analyses existing code, writes changes and scripts, prepares the tests and documents every step. Each request has its own conversation, so the work stays organised.
Part of delivered systems
In some projects Claude runs in production under written rules: it handles events in the order flow of a B2B store, answers customers on WhatsApp by searching the real catalogue and writes the hourly summary of important emails. Sensitive decisions stay with people.
A project, step by step
Who does what
| Claude | Us |
|---|---|
| analyses code, databases, logs | decide what gets done and in what order |
| writes code, scripts, firmware, documentation | install, wire and measure on site |
| tests on the test environment and places test orders | check the result before production |
| prepares the deployment package and rollback steps | answer to the client |
| works by written procedures in production systems | keep the access rights, passwords and approvals |
Fixed rules
- Nothing goes to production without a test and our confirmation.
- Every change has rollback steps and is recorded in a log.
- Client data is handled on copies wherever possible; emails from the test environment never reach clients.
- Claude is not given passwords and sends nothing on our behalf without confirmation.
- One ticket means one conversation, so two jobs never overlap on the same environment.
Claude in real projects
Four case studies where Claude's role is written into the project documentation.
B2B platform for laboratories
Each request is a ticket handled in a separate conversation. Claude analyses, writes and tests; the client starts the deployment. Newsletter: from 26.5 s to 5.3 s.
Automated B2B stores
Claude handles events in the order flow by a written procedure and answers on WhatsApp from the real catalogue. Cancellations stay with us.
AI email assistant
Every hour Claude picks out the important emails and writes draft replies. Nothing is sent without confirmation and nothing is permanently deleted.
Server infrastructure
Claude works on the server over a secure connection and reads the hardware health report every day, noting what changed since the previous day.
What clients ask us
Does Claude have access to our data?
It works on copies and test environments. Production access is controlled by us and logged.
What happens if Claude makes a mistake?
Mistakes are caught on the test environment. If something slips through, every deployment has rollback steps. We record mistakes too, not only successes.
Why not just say "we use AI"?
Because we want it to be clear who does what: Claude does a significant part of the work, and the responsibility stays with us.
An application without a developer, or a process done by hand?
Or an installation that needs to be connected to software. Write to us and we will tell you how we would approach it. See also AI assistants for business.