GPT 5.1 and Claude 4.5: Revolution in Autonomous Coding

The Emergence of New Coding Models

Last November, the world of software development experienced a major upheaval with the arrival of GPT 5.1 and Claude Opus 4.5. These models, developed in artificial intelligence research labs, have crossed a crucial threshold in terms of performance. Unlike their predecessors, they no longer just produce code that requires constant oversight. Now, they are capable of generating functional applications in a nearly autonomous manner. This advancement allows developers to request the creation of complex applications, such as a Mac application, and receive results that go beyond the stage of a simple prototype riddled with bugs.

Software Engineers on the Front Line

Software engineers find themselves at the heart of this technological revolution. Thanks to these new tools, it is now possible to produce 10,000 lines of code in a single day, with the majority of that code functioning correctly. However, the question remains: how do we move from "most works" to "everything works"? This inquiry reflects the challenges faced by engineers, who are now positioning themselves as forerunners for other information professions. Code, by its binary nature, is easier to evaluate than other complex tasks, such as essay writing or preparing legal documents. Engineers must therefore rethink their roles and ways of collaborating, as tasks that once took a lot of time are now automated.

The Mobility of Coding

The evolution of coding tools has also transformed the way developers work. It is now possible to write code on a mobile phone, an advancement that seemed unimaginable a few years ago. Thanks to the Claude app on iPhone, developers can now code on the go, whether strolling on the beach or in other settings far from their traditional office. This flexibility offers a new dimension to engineers' work, allowing them to engage in coding tasks in various environments.

Towards Responsible Coding

With the increased power of these tools, the question of responsibility in coding becomes crucial. As long as the code is used for personal purposes, the consequences of bugs are limited to the user. However, once that code is shared or used by others, the implications can be much more serious. Developers must therefore adopt a cautious and ethical approach, ensuring that their code does not pose risks to end users.

The Automation of Software Factories

The idea of a "dark factory," where processes are so automated that no human intervention is necessary, is becoming a reality in the field of software development. Today, 95% of the code produced by some developers is no longer typed manually. Current AI models can perform complex tasks, such as renaming variables or refactoring code, faster than a human could. This automation even extends to code reading, with companies like StrongDM exploring new methods to manage code without direct human intervention.

The Shift of Bottlenecks

The automation of coding has shifted the traditional bottleneck of software development to the testing phases. Previously, creating a specification and implementing it took weeks. Today, thanks to coding agents, this process can be reduced to a few hours. This allows for faster testing of ideas and the construction of functional prototypes in record time. Developers can thus explore multiple approaches for the same feature, optimizing the development process.

The Mental Exhaustion of Engineers

Despite technological advancements, the effective use of coding agents remains demanding. Experienced engineers, with 25 years of experience, find this work mentally exhausting. Managing multiple coding agents simultaneously, each working on distinct problems, can quickly lead to fatigue. It is therefore essential for developers to find a balance and set new boundaries to avoid burnout.

The Redefinition of Interruptions

The impact of interruptions on developers' work has also evolved. While it was once crucial to preserve long periods of uninterrupted focus, new working methods allow for more frequent interruptions. Developers can now interact with their coding agents in just a few minutes before moving on to other tasks. This flexibility offers a new work dynamic, quite different from traditional methods.

The Reevaluation of Time Estimates

Advancements in AI have disrupted the methods of estimating development times. With 25 years of experience, engineers had acquired expertise in assessing the time needed to complete a project. Today, these estimates are often called into question. Tasks that once took weeks can now be completed in 20 minutes thanks to AI. This uncertainty pushes developers to experiment more, launching tasks to AI to explore new possibilities.

Challenges for Intermediate Engineers

Finally, the impact of these tools varies depending on the experience level of engineers. ThoughtWorks, an IT consulting firm, recently held a seminar to discuss these issues. AI tools are particularly beneficial for experienced engineers, who see their skills amplified, and for newcomers, who benefit from easier integration. However, intermediate engineers may face difficulties, as these tools do not always meet their specific needs.