AI Will Reshape How We Build Software, Experts Say

A recent meeting of 33 experts from software engineering, artificial intelligence, and human-computer interaction has sparked a crucial conversation about the future of software development in the age of AI. Held at Shonan Meeting 222 in Tokyo, the discussions centered on how AI foundation models like GPT-4, ChatGPT, Copilot, and Code models such as llama2 and Deepseek are poised to change integrated development environments (IDEs). These models have already shown remarkable performance in tasks like code generation, testing, code review, and program repair, outperforming state-of-the-art approaches by large margins in some cases. As AI becomes more integrated into development tools, the meeting aimed to address fundamental questions about what the future IDE should look like, s and opportunities AI brings, and how to build AI agents into new environments.

The key finding from the meeting is that AI foundation models are likely to drive an evolution in software development rather than a complete revolution. Participants debated whether AI represents a radical overhaul or a step forward in existing practices, with many concluding that core tasks—design, coding, debugging, and testing—will remain unchanged. Instead, AI is expected to automate repetitive or mechanical aspects, freeing developers to focus on high-level design and creative problem-solving. This shift could streamline workflows, reduce boilerplate work, and accelerate delivery cycles, enhancing overall productivity. However, the experts emphasized that programming must stay an immersive, collaborative human activity, with developers intervening in critical moments like debugging complex issues or architecting scalable systems.

Ology involved a structured workshop format over four days, with participants engaging in breakout groups to explore different perspectives on AI's impact. Before the event, a pre-survey highlighted five key questions: which software development tasks AI should handle, how AI should be integrated into IDE features, what role remains for humans, whether development environments are still needed, and bold claims for software engineering in 2050. During the meeting, attendees were divided into four breakout groups, each focusing on a specific angle: existing tools and revolution, evolution of software development, human and process integration, and radical futuristic ideas. These groups engaged in deep discussions, walking sessions, and feedback rounds to synthesize insights, with plans to publish separate articles based on their .

Analysis from the breakout groups revealed nuanced views on AI's role. Breakout Group 1 argued that AI-augmented IDEs are an evolutionary step, moving developers to a higher level of abstraction where tedious work is handled by computing power, leaving tasks of critical thinking. Breakout Group 2 supported this by noting that past innovations like compilers and IDEs increased software engineering capacity rather than replacing it, and they debunked misconceptions such as GenAI replacing all software engineers or prompts becoming the new source code. Breakout Group 3 envisioned future IDEs as adaptive, context-aware cognitive companions that personalize interactions based on user personas, project types, and emotional states, integrating prompt-as-documentation for transparency. Breakout Group 4 explored radical ideas, suggesting that by 2050, software development and usage might blur, with natural language, gestures, or biological signals enabling continuous modification, and simulators allowing immersive what-if scenarios.

Of these discussions are significant for both developers and the broader tech industry. As AI foundation models become more embedded in IDEs, they could democratize software development, allowing domain experts and laypeople to create software through natural language without deep coding knowledge. This could lead to more efficient and personalized development environments that adapt to individual needs, potentially transforming how teams collaborate and manage projects. However, the experts caution that this evolution must balance automation with human autonomy to avoid over-reliance on AI, maintain accountability, and preserve creativity. The shift also raises questions about data management, prompt design, and the need for new engineering techniques to support AI training and deployment across diverse hardware and software platforms.

Limitations of the current understanding were highlighted throughout the meeting, pointing to areas that require further research. s include aligning training data across software engineering and AI domains, managing divergent terminologies, and ensuring that AI tools do not obscure underlying implementation details. There are also risks associated with over-personalization, privacy concerns from collecting behavioral data, and potential biases in AI systems that could privilege certain user profiles. Additionally, the experts noted that while AI excels in greenfield development, its performance in complex, established brownfield systems remains uncertain, and human coordination will still be necessary for tasks involving fundamental uncertainty. Future work should focus on developing ethical boundaries, assessing impacts on productivity and collaboration, and building transparent, trustable AI-human interactions in development environments.