From prompts to production: AI will soon write most code, reshape developer roles

At Meta’s first LlamaCon AI event last week, Microsoft CEO Satya Nadella said AI now writes up to 30% of the company’s code — and Meta CEO Mark Zuckerberg revealed that his company is developing an AI model to create future programs for its AI systems.

“Our bet is sort of that in the next year probably…, maybe half the development is going to be done by AI, as opposed to people, and then that will just kind of increase from there,” Zuckerberg said.

AI-augmented coding tools are set to revolutionize software development by creating source code, generating tests automatically, and freeing up developer time for innovation instead of debugging code. Some industry forecasts predict a 30% productivity boost from AI tools, potentially adding more than $1.5 trillion to global GDP.

One of the most popular AI-assisted coding methods now is known as “vibe coding,” or using natural language prompts (NLP) in a conversational way. Generative AI (genAI) tools are capable of offering contextual ideas and generating code based on the conversation.

By 2028, 75% of professional developers will be using vibe coding and other genAI-powered coding tools, up from less than 10% in September 2023, according to Gartner Research. And within three years, 80% of enterprises will have integrated AI-augmented testing tools into their software engineering toolchain — a significant increase from approximately 15% early last year, Gartner said.

A report from MIT Technology Review Insights found that 94% of business leaders now use genAI in software development, with 82% applying it in multiple stages — and 26% in four or more.

Some industry experts place genAI’s use in creating code much higher. “What we are finding is that we’re three to six months from a world where AI is writing 90% of the code. And then in 12 months, we may be in a world where AI is writing essentially all of the code,” Anthropic CEO Dario Amodei said in a recent report and video interview.

A real shift in software creation

While that timeline might sound bold, it points to a real shift in how software is built, with trends like vibe coding already taking off. Diego Lo Giudice, a vice president analyst at Forrester Research, said even senior developers are starting to leverage vibe as an additional tool. But he believes vibe coding and other AI-assisted development methods are currently aimed at “low hanging fruit” that frees up devs and engineers for more important and creative tasks.

A survey of more than 2,300 developers earlier this year found that 42% are already using what Forrester terms “TuringBots,” or AI-based code generators. “The opportunity is too big to ignore,” Guidice wrote in an April report, “The Architect’s Guide to TuringBots.”

“As TuringBots become smarter and more autonomous and enterprises leverage their capabilities beyond just code generation, teams will automate more concatenated software development lifecycle (SDLC) tasks and build end-to-end apps that today take weeks or months to deliver in near real time,” the report states.

AI assistants like GPT-4 Turbo (ChatGPT), GitHub Copilot, Cursor, Replit Ghostwriter, Codeium and Amazon Q Developer all support vibe coding by enabling intuitive, conversational development.

Augmented coding tools can help brainstorm, prototype, build full features, and check code for errors or security holes using natural language processing — whether through real-time suggestions (Copilot), interactive code editing (Cursor), or full-stack guidance (ChatGPT). The tools streamline coding, making them ideal for solo developers, fast prototyping, or collaborative workflows, according to Gartner.

GenAI tools include prompt-to-application tools such as StackBlitz Bolt.new, Github Spark, and Lovable, as well as AI-augmented testing tools such as BlinqIO, Diffblue, IDERA, QualityKiosk Technologies and Qyrus.

Apple is reportedly working with Anthropic to bring AI coding tools to Xcode, using the Claude Sonnet model to support AI-generated code writing and testing. The tools are being tested internally, with no confirmed plans for public release, and Apple is unlikely to comment before next month’s WWDC.

After a prewview In 2023, AWS launched Q Developer in April 2024. Developers use Amazon Q Developer to vibe code in the command line interface (CLI), and last week, AWS expanded this agentic experience to the integrated development environments (IDE) for Visual Studio Code so developers can now also vibe code in the IDE.

Srini Iragavarapu, director of GenAI Applications and Experiences at AWS, pictures developers with mood lighting and music having a conversation with AWS’s AI tools to create new products and fix code in existing code base. Iragavarapu compared AI-assisted coding to conventional “pair programming,” where two developers work together on the same project.

“It’s a very fluid interaction with the AI systems. Now the paired programmer…is the AI assistant,” he said. “These agents are task based. They are goal seeking, similar to how the programmer sitting next to you is goal seeking in that they would want to finish a task at the end of the day.”

Internally, Amazon developers all have access to the Q Developer tool set. Recently, the company used Q Developer to help it update 30,000 applications from an older version of Java to a newer version. AWS believes it saved 4,500 years in software engineering time that it would have otherwise have taken a team of engineers to complete the upgrade, as well as $260 million in annualized efficiency gains, Iragavarapu said.

A foundational change in the Software Development Life Cycle

Gartner Research, in a survey published last month, found that 35% of IT leaders expect genAI to fundamentally change their organization, with 52% expecting their organization to use the technology to build software. More than two-thirds of the execs surveyed also believe genAI benefits outweigh the technology’s risks.

Using genAI in software engineering leads to an increased focus on team productivity due to the widespread perception that the key benefit from the technology will be cost reduction, Gartner said

Armando Franco, director of Technology Modernization at IT consultancy TEKsystems Global Services, agreed most businesses have jumped on the AI train; and, while development jobs will still exist, they will be completely different. 

Overall, Franco argued, genAI tools will not be able to generate 90% of the application technology stack; more accurately, that figure could be around 60% to 70% of the application code base. That breaks down into several key areas, such as:

Simple App Code: 60-90%

API & Middleware: 50%

Data Layer: 40%

IaC: 80%

Networking: 25%

Security & Policy: 25%

Operations and Observability: 50%

Franco based his numbers on the current capabilities of available genAI tools and what he’s seen through internal development, “and it changes daily and can be very expensive depending on the models used,” he said.

“This will evolve,” Franco said. “For now, genAI can generate most of the basic, common, repeatable patterns. For advanced or complex scenarios, human-guided code development is required. As the platforms evolve, genAI will provide a platform for developers to guide genAI on how to develop more complex features.”

The same holds true for AI agents; developers can use genAI to develop complex agents that can then be deployed and trained on the same platform, he said.

Current AI tools, while powerful, are often costly to use and they change rapidly, Franco said. He described an evolution in software engineering, where engineers become specialized AI-powered architects who design and maintain complex systems with rapid, user-driven updates. Prompt engineering skills will be essential, too, as guiding AI effectively becomes a core skill.

Rather than competing with the technology, engineers will work alongside it in smaller, expert teams that build faster, higher-quality applications and drive innovation across industries, he said.

Developers find genAI tools most useful for tasks like boilerplate generation, code understanding, testing, documentation, and refactoring. But they also create risks around code quality, IP, bias, and the effort needed to guide and verify outputs, Gartner said in a report last month.

“Software engineers should assess the potential benefits and update fundamental processes to ensure the successful adoption of genAI-augmented development tools,” Matt Brasier, a Gartner vice president analyst, wrote in the report.

Even as developers increasingly rely on AI-augmentation tools, “humans must stay in the loop to understand what code is being deployed and how,” said AWS’s Iragavarapu. “Writing the code and building the application is [just] a lot easier. And in fact, as you’re deploying and debugging…understanding the logic of what it is doing and how it is doing [it] still are very relevant and prevalent.”