A new wave of coding agents promises to automate routine technical labor, transforming the tech industry workforce. However, experts question whether these agents can alleviate software development’s biggest headaches.
Microsoft recently announced an AI-powered coding agent for GitHub Copilot, designed for time-consuming but boring tasks such as adding features, fixing bugs, and improving documentation. The agent excels in well-tested codebases with low-to-medium complexity tasks.
Other tech companies, like OpenAI, are also investing in coding agents. Amazon Web Services boss Matt Garman previously suggested that human coding could disappear within two years, but later clarified that the need for human development is still essential for tasks such as documenting code and upgrading software versions.
While AI-generated code may reduce the workload of programmers, it can also introduce new errors that will become apparent when programs age or face unexpected tests. The software industry’s assumption that what works in tech will work everywhere else could be challenged when these techniques are applied to other industries.
As coding agents take over routine labor, product designers and creative engineers will use “vibe coding” – a technique that involves improvisational rough drafting via AI prompting – for fast prototyping of new ideas. However, the biggest challenges in creating software still arise from poorly conceived specifications and misinterpretations of data, which are often rooted in confusion over human needs.
Despite these concerns, developers who excel at navigating the boundaries between human desire and machine capability will remain in demand. The need for human expertise in software development is expected to endure.
Source: https://www.axios.com/2025/05/20/ai-agents-software-programming-coding