From programmer to architect: the new role of developers

Code is becoming a commodity

Anyone commissioning a digital project today encounters developers who work differently than they did three years ago. According to Google's 2025 DORA report, 90 per cent of the roughly 5,000 professionals surveyed use AI in software development. The 2025 Stack Overflow survey, with around 34,000 participants, puts the figure at 84 per cent who use AI tools or plan to. In its study „Softwarewelt 2036“, Bitkom draws the conclusion: code itself is becoming a commodity. AI agents produce code, tests and standard logic on demand.

My argument: developers are not losing their value, but their old place in the project. They move from the end of the chain to the beginning. That is where they define what gets built, how the parts fit together and who answers for the result. The programmer becomes an architect who can still lay bricks if needed.

New tasks for software developers

Gartner described the shift in 2024 in three stages. In the short term, AI tools only moderately increase productivity because they supplement existing ways of working. In the medium term, AI agents take over entire tasks that developers hand off. In the long term, a professional profile emerges that combines software engineering, data skills and AI. Gartner therefore expects 80 per cent of developers to have to upskill by 2027.

Bitkom names what will then count: goal definition, architecture, context, quality and responsibility. Concepts and an understanding of objectives weigh more heavily than syntax; specialising in individual programming languages loses value. Golo Roden, technical managing director of a software company, put it pointedly on heise: when code becomes cheap, the question that matters is which code should be written at all. A language model does not know the logic of a particular company.

I ask three questions of every project before I build: who will work with it, what becomes easier, and whether people will enjoy using it. AI changes nothing about these questions, but it makes skipping them more expensive, because the wrong code now takes minutes instead of weeks.

Responsibility stays with people

The more code machines write, the less the people working with it trust it. That has been measured. In the 2025 Stack Overflow survey, 29 per cent of developers trust the accuracy of AI results; a year earlier it was 40 per cent. Only 3 per cent trust them highly. Two thirds of respondents name the most common complaint: the solution is almost right, but not quite.

This is precisely where the new task comes from. Bitkom notes that human approval remains relevant at the appropriate points and that every agent action and every code change must be logged traceably. Responsibility stays with people. That matches my experience: I switched a solution to an AI-based version that generates mandatory documents automatically. A human checks and approves. Just as the architect answers for the structural engineering, the developer will answer for the result, even without having laid a single stone.

What speaks against this argument

The objections deserve to be taken seriously. In 2025, the research institute METR observed 16 experienced developers working on their own projects in a controlled trial. With AI tools they took 19 per cent longer, yet believed afterwards that they had been 20 per cent faster. A repeat study at the end of 2025 with 57 developers found no robust advantage, and METR itself considers this data of limited reliability because of biases in task selection. Anyone claiming that AI makes every developer twice as fast has no clean evidence for it.

The DORA report adds: AI amplifies what is already present in a team, strengths as well as weaknesses. And the Bitkom study is based on interviews with executives and, by its own account, is not representative. The study describes an expectation, not the current state of affairs. Yet these very objections support the argument. If AI merely amplifies, the result depends more than ever on those who set the goals and define the architecture.

What this means for companies

For managing directors and legal departments commissioning software, the shift has three consequences. First, billing by the hour comes under pressure. Bitkom expects clients to pay for measurable results in future, such as a closed security gap or a passed audit. When comparing offers, you are therefore better off looking less at the daily rate and more at who is responsible for the concept.

Second, who can build is changing. Bitkom sees subject-matter experts creating production-ready solutions with AI tools, not just prototypes. That is an opportunity and at the same time a liability issue, because whoever produces software within a company bears responsibility for it, whether they see themselves as a developer or not. Third, the required profile is shifting. Bitkom now describes the skills shortage as a shortage of people who combine subject expertise with the ability to steer AI. What is sought are translators between the business side, technology and regulation.