What Claude Code is
Claude Code is an AI tool for programming made by Anthropic. It works directly inside the project: it reads the existing code, writes new parts, runs tests and proposes changes. We use it to build custom business software for mid-sized companies, meaning companies with 50 to 500 employees in Austria and Germany.
This article is not written for developers but for the people who commission software: managing directors and heads of IT. It answers three questions. What does the AI actually speed up? What stays with people? And what does that mean for the timeline, quality and price of your project?
What the AI speeds up
A large share of the work in a software project is writing. User interfaces, forms, data access, interfaces to existing systems and the tests that go with them follow clear patterns. Claude Code writes these parts to our specifications in a fraction of the time. Parts that used to take days are now built in hours.
The AI saves the most time on:
- Code that follows defined patterns: screens, lists, data access and interfaces, once it is clear what they should look like.
- Automated tests: every change comes with tests that check whether the business logic is right and whether existing features still work.
- Existing code: the AI quickly reads its way into systems that have grown over years and helps to understand how they fit together before anything is changed.
- Restructuring: when a structure has to change, the AI applies the change consistently in many places, safeguarded by the existing tests.
What stays with people
The AI writes code. But it does not know how your company works, which exceptions exist or who is allowed to see which data. These questions stay with people, and they are the core of a good software project.
With us, these steps always lie with people:
- Clarifying requirements: with you, using real examples from your day-to-day work.
- Deciding what gets built: business logic, permissions and roles, data model.
- Reviewing: code review, a test run and a cross-check in the real process.
- Signing off: a change only goes live after this review.
- Taking responsibility: there is a dedicated contact for the software in operation, even years after launch.
The AI works under supervision. It writes code and tests. The decision about what goes into your software is made by a person.
The review chain: tests, code review, sign-off
AI-generated code is not automatically secure. In a study by the security vendor Veracode from July 2025, AI-generated code contained a vulnerability from the OWASP Top 10 in 45 per cent of 80 test tasks. The OWASP Top 10 is a widely used list of the most important security risks in web applications. Code from the AI is therefore only as secure as the review that follows.
That is why no change goes live with us without three steps. First, automated tests: every change comes with tests, and all existing tests have to keep passing. Second, code review: a person reads the change and checks whether it is correct, secure and consistent with the rest of the system. Third, sign-off: only once tests and review have passed and the change has been checked in the real process does it go into operation.
A window manufacturer whose software we look after shows how this holds up over years. Sales, measurement, capacity planning, service and installation, and training all run there in one shared software. It has been in daily use since 2022 and is synchronised every night with the ERP of 11 sites. Around 6,000 automated tests protect it. In 2026, four large modules were built there within five months, and a third-party product the company had bought in was replaced.
What this means for timelines and fixed prices
The time the AI saves on writing does not simply disappear. It goes into alignment with you, into tests and into review, which are exactly the parts that decide how the software performs in operation. For you, this has two consequences.
First, the timeline gets shorter. After the analysis, a small business application is often in use within four to eight weeks. Larger systems are built in stages of a few weeks each, so that the first parts can be used early.
Second, the price becomes predictable. We do not charge by the hour but at a fixed price per stage. The fixed price follows a paid analysis lasting one to two weeks. It costs 3,000 to 6,000 euros net and is fully credited when you place the order. If there is no usable result, you get the fee back. As a guide, all amounts net: a small business application usually costs between 10,000 and 30,000 euros, a system for several departments between 30,000 and 90,000 euros, and a company-wide system starts at 90,000 euros.
What to look for in a provider
Many providers advertise with AI today. For you as the client, what matters is not whether a provider uses AI. What matters is how they safeguard the result and who answers for it.
These questions help you assess a provider:
- Who reviews the code the AI writes, and how is that documented?
- Are there automated tests for every change, and do they run before every sign-off?
- Who owns the source code, and is that set out in the contract?
- Where does the software run, and who keeps authority over infrastructure and data?
- Who is responsible for operation and maintenance after launch?
With us, the source code belongs to you. The software runs on your own premises or in a data centre in Austria. Operation and maintenance with updates, monitoring and a dedicated contact start at 300 euros a month.
Code from the AI is only as secure as the review that follows. That is why a person decides what goes live.
If you are planning a software project and would like to know what it would look like with this way of working, we are happy to talk it through. The initial call takes 30 minutes and is free of charge. Afterwards you will know whether custom software is worthwhile for you and roughly what it would cost.