Artificial intelligence-based tools, so-called „coding agents“, are fundamentally changing the way software is created today. But not all areas of development are accelerating in the same way. Understanding these differences is key to getting teams right and setting realistic expectations.

The biggest contribution of AI is seen in frontend development. For example, when creating websites for e-shops. Modern models have an excellent command of languages such as TypeScript or JavaScript and frameworks such as React or Angular. Thanks to this, they can generate functional interfaces very quickly and, in addition, iterate independently based on the results in the browser. Visual design remains a weakness for now, but if the design is predetermined, implementation is significantly faster than before.

Backend development, such as creating APIs to work with data, is more complex. It requires more careful guidance from experienced developers, especially because of edge cases that can lead to bugs or security issues. In addition, errors in the backend often cause opaque problems such as database corruption or incorrect query results. While AI helps speed up the work, it is still true that a quality backend requires experience and thorough design.

AI has an even smaller contribution to make in infrastructure. Tasks like scaling a system for thousands of users while maintaining high reliability require deep knowledge and a sense of trade-offs. Current models have limited knowledge in this area and are therefore not very reliable in making critical decisions. Moreover, detecting infrastructure errors, such as network misconfigurations, is one of the most challenging tasks ever.

So far, AI has been the least helpful in research. This involves formulating hypotheses, experimenting, interpreting results and testing them repeatedly. While coding agents can speed up the writing of code for experiments or help to organise them, most research work involves thinking and decision-making, where AI has a limited role so far.

The division of software work into frontend, backend, infrastructure and research is simplistic but useful. It helps companies better organize teams and set expectations. While frontend teams may work significantly faster today than they did a year ago, the pace has changed only minimally for research teams.

Software development is thus entering a new phase where speed increasingly depends on the right use of AI - but also on where its limits are not yet exceeded.

deeplearning.ai/gnews.cz - GH