Fully Automated Software Development: Will Programmers Disappear?

Originally published in Chinese on HK01 on 2025-09-04 07:00 | By Michael C.S. So | AiX Society

As a columnist, I often ponder a question: if I have several different business ideas in mind, what would it look like if they could be operated entirely through AI automation? Imagine no longer needing to go back and forth with software engineers, no more waiting through lengthy development cycles, and no more headaches over debugging or maintenance. If AI could fully automate the development and deployment of applications, it would fundamentally transform the way we start businesses and innovate. This brings us to a core question: is “fully automated software development” truly possible? And if so, when will it become a reality?

From “Writing Code” to “Talking to AI”

This future is already taking shape. Generative AI tools such as GitHub Copilot, ChatGPT, and Claude can automatically generate code, fix bugs, and even complete simple applications. According to GitHub’s research, developers using Copilot see coding speed improvements of up to 55%, with approximately 30% of suggestions being directly accepted, and 90% of developers having already submitted code that includes Copilot-generated content. By 2025, Microsoft announced that Copilot had reached 20 million users and launched a “Coding Agent” mode capable of executing multi-step tasks, enabling AI to handle more complex development work. This already signals that software development is shifting from “humans writing code” to “AI leading, engineers supervising.”

On another front, “low-code/no-code” platforms allow people without technical backgrounds to drag and drop modules, enter simple instructions, and build business processes or mobile applications. This trend is accelerating the spread of digital innovation, further breaking down technical barriers.

At the same time, a new generation of “AI Dev Agents” — such as Devin by Cognition — has demonstrated the ability to self-plan and self-debug: it can automatically write code, run tests, and iteratively optimize. The company secured $500 million in funding in early 2025 and gained adoption by major enterprises including Goldman Sachs, Ramp, and Nubank, reflecting strong confidence from both capital markets and industry in these automation tools.

Why Full Automation Remains Elusive

Despite AI’s remarkable progress, three major hurdles stand in the way of full automation:

First, the Specification Gap. Human requirements are often vague or contradictory, and they can change at any point during development. While AI can generate code, it may not truly understand the underlying business logic or market intent.

Second, Security and Reliability. Software must not only function — it must meet security, compliance, privacy, and performance requirements. Today’s AI still “hallucinates,” producing incorrect or incomplete code. In high-risk industries such as healthcare and finance, this level of risk is unacceptable.

Third, Social Integration. Software must be embedded in the real world, involving law, culture, and user experience. AI can mimic norms, but it struggles to truly understand user emotions, social conventions, or the complex relationships within organizations.

Industry Perspectives and Timelines

Regarding the future of automated programming, many tech leaders have spoken publicly. NVIDIA CEO Jensen Huang has stated on multiple occasions: “The programming language of the future will be human language,” and even emphasized that “everyone is now a programmer — that is the miracle of AI.” Microsoft CEO Satya Nadella has declared that AI is “compressing the changes of the past thirty years into the next three,” and previewed how AI agents will redefine SaaS and knowledge work, fundamentally changing the way applications are developed.

Research institutions have also provided concrete timelines. Gartner predicts that by 2028, one-third of enterprise software will have built-in AI agents, and approximately 15% of daily decisions will be made automatically by systems. McKinsey estimates that by 2035, automation and generative AI could affect up to 55% of work activities. These data points and forecasts provide clear evidence that 2025–2035 will be the critical decade for automated software development.

The Capability Curve Breakthrough

Technical benchmarks also reveal leapfrog progress. Take AI model performance on SWE-bench (a software engineering bug-fixing benchmark) as an example: in early 2023, the best model could solve only 2% of problems; in 2024, Devin pushed the success rate to 13.86%; and by 2025, OpenAI’s GPT-4.1 could solve 54.6% of tasks, with GPT-5 reportedly reaching 74.9%. While these figures remain debatable, the overall curve clearly shows that AI has moved from “barely able to handle anything” into the territory of “completing more than half of tasks” — approaching the tipping point of large-scale practical deployment.

The Real Challenge: Imagination and Governance

As “writing code” is progressively taken over by AI, the greatest challenge is no longer technical — it is “how humans will use this capability.”

In the future, anyone will be able to generate an application with a single sentence. On one hand, this could lead to an overflow of “junk apps,” flooding the market with low-quality software. On the other hand, it could usher in a new golden age where truly creative individuals can instantly bring their ideas to life, driving entrepreneurship and innovation at unprecedented speed.

Therefore, future competition will no longer be about who has the best technology, but rather:

• Whose imagination holds greater value — who can propose ideas with the most significant social and commercial impact.
• Who can establish effective governance and ethical frameworks to determine which applications should exist and which should be prohibited.

This is not merely a challenge for businesses — it is one that governments and society must confront together.

Within a Decade, Definitions Will Be Rewritten

Drawing together the statements of tech leaders, data from research institutions, and the current state of the industry, we can arrive at a clear conclusion: 2025 to 2035 will be the pivotal decade in which software development transitions from “primarily human-driven” to “AI-led.”

Jensen Huang declared that “human language will be the programming language of the future.” Satya Nadella emphasized that “AI agents will reshape application development.” Gartner forecasts that by 2028, one-third of enterprise software will embed AI agents. McKinsey projects that by 2035, more than half of work activities will be affected by automation. Combined with Copilot’s widespread adoption, Devin’s massive funding rounds, and the SWE-bench capability curve breakthrough, all of these confirm that the future of automated software development is already upon us.

However, this does not mean programmers will vanish entirely. The form of programming will be redefined, and the human role will shift upward to higher levels — planning specifications, designing architectures, ensuring compliance, and setting ethical and social boundaries.

When that time comes, the real question will not be “do we still need programmers?” but rather “how much control over shaping the world should humans retain?”

In this sense, programmers may gradually recede into the background, but the role of the “creator” will never disappear.