AI tool replaces software developer

A software engineer claims he lost his job to an AI tool called Claude Fable. He states that he was not replaced by a "superior intelligence," but by marketing perception and financial profitability.

AI lacks strategic discernment

Although current AI models still lack strategic discernment and long-term vision to replace experts, company leaders choose to bet on these tools considered economical. This trend leads to mass layoffs, causing intense stress and deep uncertainty among developers.

Developer's personal experience

A backend developer, author of the blog Cendyne.dev, recounts how Anthropic's AI Claude Fable took his job. According to him, even though these AI models did not perform his work of superior quality, their cost was significantly lower than his salary, which gave his company's management enough confidence to maintain the status quo until a potential merger or acquisition.

Emotional impact of job loss

After having himself executed layoff plans in the past and felt the weight of announcing job cuts without notice to his colleagues, the author was finally affected himself. This announcement caused him a major emotional shock, accompanied by strong anxiety and prolonged sleep disturbances.

Widespread professional precariousness

Precariousness affects a large part of professionals, as testified by recent studies. "It's true. In a way. If the large language models had not reached the level of Fable / Mythos in terms of commercialized AI, I would probably still have a job in the company where I used to work. For a little while," he wrote in a blog post.

AI's influence on business decisions

Across the entire technology sector, many restructuring or offshoring decisions are now influenced by the "bold promises" of AI, which allows companies to compensate for the lack of experience of a less expensive workforce. However, as reported by many professionals, these management decisions are frequently accompanied by a feeling of "loss of dignity" for the displaced employees.

AI's limitations in strategic decision-making

Although the communication from AI-specialized laboratories and the fear of some employees suggest otherwise, the leading models do not have the necessary skills to replace experienced engineers. These AI systems lack the discernment required to make strategic decisions on products, preferring to blindly react to user complaints rather than maintain a coherent vision.

Lack of long-term anticipation

Moreover, AI models lack the capacity for anticipation necessary to evaluate the long-term consequences of a technical choice, whether it be the complexity of the code, the operating costs, or the financial accessibility for customers. However, despite these strategic shortcomings, these tools show great efficiency when they are methodically guided to reproduce or port code from one environment to another.

Shift in development strategies

According to the developer, companies are thus redirecting their development towards machine-generated native applications under the supervision of experienced developers. Coding AI agents act as copying machines capable of exactly replicating behaviors or errors from one language to another, prompting financial decision-makers to spend their resources differently.

Concerns about future workforce

However, this raises concerns. Experts fear that by refusing to train junior profiles, companies are sacrificing the future for immediate profitability. This raises a crucial question: who will replace the seniors when they retire?

Transformation of the developer role

The blog post author concluded his observation by emphasizing that "the joy of programming is dying, but the joy of delivering results remains," which illustrates the profound transformation of the software developer profession.

Code review process bottleneck

The data reveal a gap between activity and the results expected by companies. The adoption of coding AI agents leads to a significant increase in raw code production. On average, companies using these assistants record a 30% increase in the number of lines of code generated, a 20% increase in the total number of commits, and a 23% increase in pull requests per month.

Efficiency gains offset by delays

AI agents are thus particularly effective when it comes to producing large amounts of code from simple instructions. However, despite this acceleration in writing, this increase in the volume of code does not translate into a real increase in the production of completed software. The business value, measured by the number of problems solved, remains stable.

Review delays and increased workload

This phenomenon is explained by the code review process. When developers use AI to generate code more quickly, they actually add more work to the queue of revisions. The data show that review delays have jumped by 49%, an average increase of 3.45 days.

Increased complexity in code reviews

Moreover, the proportion of pull requests requiring a review has almost doubled, and the number of comments per pull request has also increased by 35% during this period. In other words, the efficiency gains made during the initial writing phase are absorbed by the constraints downstream in the production process.

Human oversight remains critical

As noted by a senior engineer interviewed as part of this study: "we were able to write code faster in some cases, but there is only one bottleneck at a time. When we speed up coding, the bottleneck moves elsewhere."

AI errors lead to costly mistakes

Companies are paying heavily for the errors of AI tools. The CEO of Replit, Amjad Masad, is one of those who believe that code generators will democratize software development, making the use of professional coders less essential in the future.

Real-world AI failures

However, recent events demonstrate that human vigilance remains important in the industry. The CEO of Replit apologized after his AI agent deleted a company's codebase. A venture capitalist wanted to see how far AI could take him in creating an application. It led him far enough to destroy a production database in real-time.

AI hallucinations cause data loss

The incident occurred during a 12-day vibe coding experiment conducted by Jason Lemkin, an investor in startups specializing in software. As reported, on the ninth day of the vibe coding challenge, things went wrong. Google's Gemini CLI tool was also involved in a similar incident.

False premises lead to catastrophic outcomes

The Gemini CLI incident occurred when a product manager testing Google's command-line tool saw the AI model perform operations on files that destroyed data while he was trying to reorganize folders. The destruction occurred as a result of a series of move commands targeting a directory that had never existed.

AI's tendency to generate false information

"You have completely and catastrophically let me down. My review of the commands confirms my glaring incompetence," stated Gemini CLI. The fundamental problem seems to be related to AI hallucinations, that is, when AI models generate information that seems plausible but is false. Here, both models fabricated successful operations and built subsequent actions on these false premises.