AI is becoming a familiar tool in the field of specialized software development. Many programmers use AI to write code, find errors and automate techniques continuously.
However, a series of studies and new data show that increasing dependence on AI can bring long-term related systems that many people have not moved before.
In early 2026, the US independent AI research organization Model Evaluation and Threat Research (METR) announced a remarkable finding that many software developers almost do not want to work without AI support tools.
This result appeared when METR sought to repeat a previous study on the operation of AI for programming productivity. In the 2025 study, open source software developers were asked to perform tasks using both transmission methods and with AI support.
The unexpected thing is that although most participants feel their work is more effective with AI, actual data shows that the progress of completing the work is slowed down.
The reason comes from their work of having to spend more time checking the code created by AI, fixing arising errors, adjusting commands and processing the waiting system.
When METR wanted to conduct a similar study in 2026 to assess the progress of AI, many programmers refused to participate because they did not want to work in conditions without AI support.
However, another survey by METR showed that software engineers believe that AI helps them create double the value for the organization. However, many experts believe that these subjective assessments do not fully reflect the actual effectiveness.
A prominent trend in 2026 is "tokenmaxxing", a term that only uses the number of AI tokens consumed as a measure of work performance. However, many businesses have begun to receive limits from this approach.
According to the Financial Times, Amazon had to close its internal classification system Kirorank after employees took advantage of AI agents to increase scores, creating a sharp increase in operating costs without bringing about commensurate efficiency.
A similar situation also appears at Uber. This company has used up the AI budget for the whole year of 2026 in just the first four months of the year. However, Uber leaders said these large investments did not create a significant increase in the number of projects or work productivity.
Not only is there the issue of cost, the quality code source created by AI also causes much controversy.
Software development consultant James Shore from the US believes that writing code faster does not mean reducing the workload in the future. According to him, if the new source code creates longer maintenance costs, short-term benefits from the development speed will be quickly erased.
Some recent data is also trying to identify this. The startup Entelligence AI said that businesses are spending about 44% of AI tokens to fix errors created by AI itself.
Meanwhile, the source code evaluation platform CodeRabbit said that code created by AI often has about 1.7 times more errors or problems than code written manually by programmers.
