AI automates code and redistributes roles: how developers and tech leads adapt to this new situation.
As with other professions, AI strongly impacts the work of developers. Tools like GitHub Copilot, Cursor, and Amazon Q Developer help generate code easily. To the point that some people wonder if the job of a developer now resembles that of a code cleaner. To find out more, we interviewed junior developers, senior developers and tech leads. Three profiles differently impacted by AI.
Junior developers: “a great servant but a terrible master.”
Junior profiles are particularly impacted by the AI surge. This seems better than young profiles and is closer to the senior level. It can generate feature implementations, perform automated code reviews and extract key information from technical documentation. It also costs less and does not require long and costly training. Even if she often shows the same limitations as junior developers: missteps, lack of discernment, or even errors in cybersecurity.
For their part, the junior developers surveyed are reluctant to spend their time cleaning up code generated by AI. Firstly, through their skills. Verifications often require a solid background. For Paul Lacoste, junior developer at MN Santé: “AI tends to propose complicated/complex solutions that solve the problem posed at a given moment. Also, junior developers prefer to use AI as a very good teacher for technical explanations. For example, concerning requests for explanations of code tools.
In addition, the generation of code by AI can lead them to develop a certain “laziness” according to Thom Costuas, junior web developer at Inodia. Which can be feared. “This leads to delegating the entire thinking and design phase to AI,” analyzes Thom Costuas. This can then become, according to the expression “a great servant but a terrible master.”
But, despite these good intentions, junior developers say they are often faced with verifying the generated code. As companies’ use of AI is important for creating code, the control phase remains necessary. And this, even in projects “without logic or reflection dependent on the environment”, for Thom Costuas. Verification can also take place with the retrieved code. “The projects on which I work or which I take over sometimes have traces of AI,” points out Thom Costuas. “Extensive comments to explain each variable or calculation stifle the code, making it more cumbersome to read or evolve. Sometimes, the nomenclatures or simply the programming logic do not fit into that of the project.”
Senior developers: adaptability as a key word
Faced with AI, the skills of senior developers must allow them to intervene quickly on different types of projects. Particularly those of importance, where humans still retain their place. For example, their knowledge of React should allow them to better understand certain problems.
This is not without consequences on their daily lives. For example, at Amazon, AI transforms thinking work into assembly line work. The time allowed for coding is drastically reduced. The role of the developer is then more to clarify and correct the code written by the AI. Fabien Dalla-Valle, software engineer at Metadot, confirms: “The transformation is felt in terms of speed. The code generated by AI makes it possible to write lines of code more quickly than before. This also changes the way of thinking about problems. Even when we know a solution, the AI can find other alternatives which may be interesting. Development is therefore faster but control and verification take more time.”
The senior developer must above all develop his adaptability, with increased technological monitoring for example. Specialized profiles are therefore on the rise in sectors such as health, finance, security and cloud architecture.
Tech leads: “While AI streamlines execution, it does not replace thinking.”
Tech leads seem less affected by checking AI-generated code. Indeed, for them, AI especially helps in the decision-making phase. “It speeds up certain tasks, especially those around the product,” says Julie Droin, head of product at Eskimoz. “We document faster, we summarize meetings and user feedback better, and we can generate code for simple functions.” So, according to her, AI does not really change the core business. “The fundamentals remain the same: ability to structure one’s thinking, to understand technical constraints, and to maintain a critical mind. On complex decisions, product framing or software architecture, human expertise remains central. If AI streamlines execution, it does not replace reflection.”
However, the verification of generated code can take place, and again be relatively time-consuming. “Sometimes I spend more time correcting or reviewing what the AI has produced than coding directly,” laments Youssef Messaoud, tech lead at Smile. “Control remains essential: the AI goes quickly, but it can also make very subtle errors. But with practice and good prompts, this improves.”
The work upstream, rather than downstream, during verification is emphasized by Julie Droin. According to her, for AI to be truly useful, it must be piloted correctly. This requires a real intellectual effort upstream. “Prompted well, AI becomes a powerful tool. Poorly prompted, it produces noise. Building a clear request, setting constraints, providing context, reformulating several times: all this requires a structuring capacity that not everyone has. AI values people who know how to think clearly and immediately exposes those who do not master this skill. It is not a magic button.”
