The state of AI in
software development

By Aatu Väisänen

March 14th, 2025

November 2022 was a pivotal point for the future of software development, as ChatGPT was released to the public. It was the real starting shot for the generative artificial intelligence (AI) boom driven by the outburst of large language models (LLMs), which continues to evolve as increasingly advanced AI models are released at a steady pace.

The adoption of AI varies significantly across different professions. A recent study by Anthropic aimed to identify early indicators of which economic sectors might experience the greatest impact as AI technology progresses, by analyzing millions of real discussions with Claude (an AI assistant created by Anthropic). According to the study, by far the largest AI adoption rate is in occupations related to software development, as 37.2% of all Claude queries related to tasks like software modification and code debugging.

It’s evident by now that software development stands out as an industry with significant potential for enhancing productivity through greater AI adoption. However, what types of tools are available to programmers for their everyday tasks?

Code generation and auto-completion

One of the most helpful AI tools for programmers to deploy in their daily work are code generation and completion tools, such as GitHub Copilot and Tabnine. Using a tool like Copilot assists programmers by providing intelligent suggestions on how to complete the lines of code that they are typing. They can also automatically complete repetitive chores, like writing boilerplate code and documentation. The biggest benefit is perhaps the fact that the tool is incorporated to the code editor, which makes it easier to stay focused to the task at hand.

There are also IDEs like Cursor, which are solely built around the idea of making the integration of AI to the daily workflows of software developers as seamless as possible.

Prototyping and design-to-code

For front end developers, getting a project bootstrapped or drawing the basic wire frame for a user interface becomes effortless with AI tools such as v0 by Vercel. The tool is constantly evolving, allowing for more complex ideas to be manifested by iterating the project by prompting the generative chat of v0.

The very essence of front end development is turning a bunch of design files to a functional user interface that looks exactly like the designer envisioned it: traditionally this has been a task that requires great knowledge of HTML, CSS, JavaScript, and a sharp eye for getting each detail just right. Like with many other developer tasks, the rise of generative AI is revolutionizing this. AI tools like Visual Copilot (by Builder.io), Locofy, and CodeParrot AI allow turning design files from Figma into functioning apps and websites without manually writing a single line of code. Of course these tools are far from perfect, and it takes some trial-and-error with prompting the AI to get every detail done right, but I can only imagine how sophisticated these tools get over time, as the technologies behind them progress even further.

Coding agents

Arguably the hottest topic around AI and LLMs has recently been agentic AI: it refers to AI systems that have been engineered to operate autonomously and proactively, taking the initiative in making decisions and taking actions on behalf of the user. The only task left for the user is to specify the goal that they want to achieve, the agentic AI system does everything else.

While agentic AI systems are emerging for generic purposes, such as conducting daily user tasks directly in the web browser (OpenAI’s Operator), in my opinion agentic AI is a real game changer in specific tasks like programming. At the time of writing this (March 2025), Microsoft’s GitHub Copilot has just recently released the preview version of their Copilot agent mode, which is available to all VS Code Insiders users. Copilot agent mode functions as an independent programming collaborator, executing tasks like codebase analysis, file processing, code edit suggestions, and performing terminal commands and tests. It can analyze and auto-correct errors in a loop until the given task is completed – all the user has to do is give commands and monitor the output that the system suggests.

A simple example of the workflow with an GitHub Copilot agent mode.

I can easily see how agentic AI might just be the logical path forward for software development as an industry: the potential impact on the productivity of programmers could be comparable to the increase in productivity when Excel revolutionized the way data management is conducted. However, it is good to keep in mind that agentic AI tools are really new, so only time will tell how big of a role they will have in the daily work of software developers around the globe.

Ethical considerations

Although various AI tools in software development are transforming industry practices and significantly boosting developer productivity, ethical considerations surrounding AI remain critical. Concerns such as code ownership, biases embedded in AI-generated code, and potential security risks tied to such code require careful evaluation by companies deciding how to integrate AI into their processes. This underscores the need for human oversight of AI-generated code. Rather than replacing human developers, AI serves as a cooperative partner in the development workflow.

Conclusion

Today we have numerous AI tools that can have a massive boost on the productivity of software developers. New code generation, prototyping, and design-to-code systems are emerging on a regular basis. Progress is tremendously fast, as the software development AI revolution has only just started. In my view, agentic AI appears to be the most promising direction for the future of software development, but as the systems mature in the following years, only time will tell what the real impact of agentic AI will be. However, while numerous AI tools are quickly disrupting industry practices, ethical concerns surrounding AI should not be overlooked when businesses continue to integrate AI into their software development workflows.