The Impact of AI on Software Development: A 24-Month Forecast

The rapid evolution of AI is positioned to fundamentally transform the landscape of software development over the next 24 months. As AI-driven tools like GitHub Copilot and Cursor AI become integral parts of the development process, they are not only enhancing productivity but also reshaping the roles of developers themselves.

By understanding these trends, we can better anticipate the profound changes that lie ahead in the software development industry.

AI Assistants and GitHub Copilot: Transforming Software Development

The introduction and dominance of GitHub Copilot with GPT-4 (announced in March 2023) has revolutionized software development by allowing developers to focus more on creativity and problem-solving rather than repetitive coding tasks.

In an experiment and study done by GitHub in May 2024, developers using GitHub Copilot completed tasks 55% faster than developers not using AI assist tools (research can be found here).

GitHub Copilot has become a co-pilot for developers, automating code suggestions and enabling them to speed up coding without context-switching between documentation and the editor. Copilot’s impact is evident in its ability to enhance productivity and optimize the software development lifecycle by reducing the need for extensive debugging and manual code writing.

However, as software development has progressed, new players have begun to disrupt the market. One of the recent key disruptors is Cursor AI with its Claude 3.5 Sonnet model, which has changed the landscape by providing even faster and more contextually aware assistance.

Unlike Copilot, Cursor is praised for its ability to handle larger contexts and integrate more effectively with codebases. Andrej Karpathy, a notable figure in AI and ML, expressed his support for Cursor AI, citing its superior performance over Copilot in terms of speed and flexibility. He pointed out that while GitHub Copilot was groundbreaking, tools like Cursor are leading in a new era of developer interaction with AI.

In this new era, programming has become more about giving instructions in plain language and refining the generated code rather than writing from scratch.

Recent Advancements and SWE-Bench Impact

In recent months, one of the most significant advancements has been the usage of the SWE-bench (Software Engineer – Bench) framework. It benchmarks AI capabilities in real-world software engineering problems.

SWE-bench tests AI models on actual GitHub issues and pull requests, representing a realistic setting for evaluating the models’ ability to solve complex code issues.

SWE-bench revealed that on October 2023, even state-of-the-art model (back than) like Claude 2, solved only a fraction of real-world issues? - Claude 2 achieved a resolve rate of 4.4% on SWE-bench, while it took less than an hour for an experienced software engineer to achieve 78%. This was pushing the industry towards more intelligent, adaptable, and autonomous AI solutions.

The new race to meet standards like SWE-bench demonstrates the emerging capabilities of AI in dealing with complex software engineering challenges by different models and vendors.?

And as can be seen on the September 2024 SWE-bench Verified Leaderboard table below, the race is on!

The latest AI models and tools (Amazon Q, GPT4o and Claude 3.5 Sonnet) are now at the top of the chart, already reaching 33%-38% resolve rate:

AI now can understand and address issues across multiple files, handle long-context programming challenges, and process large codebases with a focus on testing and debugging.

With the current pace of model progression, we can expect AI to reach a level of senior software developers in SWE-bench scale within 12 months.

The Future of Software Engineering and Hiring Processes

With these rapid advancements in AI tools, the role of a software engineer will fundamentally change over the next two years. Instead of manual code generation, engineers will increasingly focus on supervising AI-generated solutions, debugging AI-generated code, and aligning to best practices for AI usage in software development.

AI will become a collaborative partner rather than just a tool, requiring engineers to focus on higher-order thinking, architecture design, and integration.

The introduction of GPT-agents (AI entities autonomously executing tasks, understanding context, and learning from interaction) is strongly supporting this trend. Instead of manually crafting every part of a codebase, developers could delegate specific tasks to these agents, allowing them to focus on high-level design and decision-making.

The ability to guide, supervise, and collaborate with GPT-agents will become a critical competency, like today's expectations for proficiency with development frameworks or cloud platforms.

In terms of hiring processes, we will continue to see strong demand for experienced developers, who will also get the option of new career paths due to the opportunities arising from the AI advancements.

Companies will need to prioritize not just coding skills but also a candidate’s ability to collaborate with AI systems. The traditional technical interview focused on solving coding problems manually may evolve into scenarios where engineers are asked to direct AI to solve problems or assess the quality of AI-generated code.

This shift means that companies will value engineers who understand how to work efficiently with AI, integrate AI-generated solutions into existing systems, and provide the necessary supervision to ensure the AI’s output aligns with business goals.

The Evolution of Programming Languages

In the beginning of 2023, Andrej Karpathy tweeted the following:

?

Programming languages have long been the medium through which humans communicate instructions to computers. However, AI’s increasing ability to understand and generate code is transforming this dynamic.

In the future, communication with computers will likely become even more intuitive, with less reliance on strict programming syntax and more emphasis on natural language instructions.?

Since this tweet, tools like GitHub Copilot, Amazon Q and Cursor AI have evolved further and gained capabilities that, using natural English, some coding tasks can be naturally performed.

The latest “Claude Artifact” with its advanced natural language processing capabilities, takes it another step further and allows developers to interact with code in a more intuitive manner, reducing the need for specialized knowledge in certain programming languages or frameworks.

We may witness a shift where AI can interpret plain English commands and convert them into optimized code, further reducing the barrier between non-technical individuals and software development.

This evolution raises the question: Will programming languages as we know them become obsolete?

- No, but.

Programming languages will still serve as the foundation for computer interaction.?

The promising future of such tools and advancements will face the limitation that natural English lacks the precision and specific syntax required to handle the logic and constraints often needed in complex applications or infrastructure development.

BUT we can expect AI to act as a mediator, allowing non-programmers to harness the power of coding through conversational interactions.

The Role of Young Generations in Predicting Technology's Future

Like other instances where rapid technological advancements have been introduced to the market, observing how younger generations interact with new technologies can often provide insights into future trajectory.

A UK 2015 academic study (by Fister-Gale, 2015), demonstrated that Gen Z readily embraced emerging technologies and expected future electronic processes to be widely available, offering consumers greater autonomy.

Gen Z's early adoption to smartphone and touchscreen technology in the early 2010s, brought companies like McDonald’s to replace many human cashiers with large self-service tablets - resulting not in a negative impact on sales, but rather in a 6% increase.

Now, consider the video below of an 8-year-old girl independently coding a Harry Potter chatbot using Cursor AI. This example provides a glimpse into what this kind of technology could mean for the software development market in just a couple of years, given the current pace of technological advancement.

Conclusion: Navigating the AI-Driven Future of Software Development

The next 24 months will be a transformative period for software development, as AI-driven tools evolve from supplementary aids to essential collaborators in the coding process. As AI continues to advance, developers will shift their roles from manual coders to strategic architects and supervisors of AI-generated solutions.

This transition will not only enhance productivity but also demand new skills, such as proficiency in AI collaboration and the ability to guide AI in producing code that aligns with complex business needs.

The impact of AI on software engineering extends beyond just the development process; it is reshaping the very fabric of how we approach problem-solving, team collaboration, and even the hiring process. AI tools like GitHub Copilot, Cursor AI, and the emerging GPT-agents signal a shift towards a future where software development is more about high-level design and strategic oversight than traditional coding.

The evolution of programming languages, influenced by AI's ability to process natural language, will democratize coding, making it more accessible to non-technical individuals. This raises important questions about the future role of programming languages and the skills that will be valued in the tech industry.

As we move into this AI-driven future, companies must adapt by prioritizing AI collaboration skills in their hiring processes and by rethinking their approach to software development.

The ability to work effectively with AI will soon become as critical as knowing how to code, if not more so.

The future of software development is bright, but it requires a proactive approach to navigate the changes that AI will bring. Those who adapt and evolve with these technologies will not only survive but thrive in this new era of software engineering.