Will ChatGPT (and others) really replace developers?

In recent months, ChatGPT and generative AI tools have captured the attention of many in the technology world, bringing back with force a question that has circulated ever since the first code generation tools appeared: will these technologies replace programmers? In our team we decided to put these tools to the test to understand whether they could truly represent a threat to the developer’s profession, and the result was a mix of surprises, disappointments, and some interesting insights.

The first thing we wanted to test was ChatGPT’s ability to generate useful code. We asked it to write scripts and functions for some of our internal projects, but the results were disappointing: the code produced was often incomplete, with errors or simply unsuitable for our needs. In the best case, ChatGPT generated educational examples, similar to those we might use in a training course to illustrate a concept without revealing the complete solution. For instance, we asked it to create a function to process data from a specific API and the resulting code, theoretically correct, lacked optimization and did not account for some project constraints. In the end we spent more time correcting that code than we would have spent writing it from scratch.

One of the areas where ChatGPT showed some potential is the generation of regular expressions: we asked it to create regex for standard and custom problems, and the results were often correct. We are not sure, however, that it represents an improvement over traditional regex generation tools, and the difference is not sufficient to justify a change in workflow.

To test its limits, we made some more creative requests. We asked ChatGPT for the last digit of pi and, with confidence, it answered that it was 9, only to then show us an example with a dozen digits that ended with a 6. We also asked it to write fantasy stories and imaginary interviews with historical figures, obtaining grammatically correct but boring and superficial results, as if written by a bored student. Finally, we asked for instructions to start a vintage motorcycle: it provided almost correct guidance while completely overlooking the crucial details to avoid getting hurt, while a colleague received the recommendation to apply maximum engine power to land a glider. It is good at generating theoretically correct content, but often lacks depth, precision, and practical sense.

Where it surprised us

The only area where ChatGPT excelled was recipe generation. We asked for a recipe for a lentil and potato stew and the result was surprisingly decent: when asked for a printable format, it provided exactly what we needed, with ingredients, quantities, and steps well structured. It also handled contradictory requests well, such as a vegan stew with chicken, politely refusing to produce an impossible recipe. It could be a useful kitchen assistant, but it is not yet ready to replace an experienced chef.

One of the most interesting questions is whether ChatGPT can become a valid assistant for programmers, helping us solve the boring and repetitive parts of the work, but our experience suggests that we are not there yet. ChatGPT cannot understand business context, budget constraints, timelines, or technological preferences, nor can it interview users and stakeholders to design solutions aligned with the business vision.

When should we start worrying?

The real question is: when will ChatGPT, or similar tools, become a concrete threat to programmers? In our view, we should start worrying when AI is able to understand business context in its entirety, including budget, timelines, infrastructure, and regulations, to design complex solutions considering both functional and non-functional requirements, and to collaborate with users and stakeholders to define objectives and priorities. In other words, when it manages to solve creative problems going beyond the simple generation of code or text. Future versions will likely be more advanced and capable of handling more complex tasks, but even then programmers will have a crucial role in guiding, supervising, and optimizing AI work.

Rather than fearing artificial intelligence, we should see it as an opportunity to improve our work and focus on more creative and strategic activities. ChatGPT is an impressive tool, but it is not yet able to replace programmers: for now it remains a useful assistant for simple and repetitive tasks, but lacks the depth, creativity, and contextual understanding needed to tackle complex problems.