10 ways generative AI will transform software development

Coding in the 1990s usually meant selecting an editor, checking code into CVS (concurrent version system) or SVN (Apache Subversion) code repositories, and then compiling code into executables. Integrated development environments (IDEs) like Eclipse and Visual Studio improved productivity by including coding, development, documentation, building, testing, deploying, and other steps in the software development life cycle (SDLC). Cloud computing and DevSecOps automation tools brought in the next wave of developer capabilities, making it easier for more organizations to develop, deploy, and maintain applications.  

Generative AI is the catalyst for the next paradigm shift. It promises to change how organizations create and maintain software and to enable new development tools and paradigms. The question for many developers and IT leaders is whether generative AI means the demise of coding as we know it. A related question is how it will affect the evolution of SDLC and DevSecOps over the next decade. With these two questions in mind, I went searching for ideas and predictions.

Is generative AI a new tool or a new way of developing? 

“I am a big believer in code, and I have seen many people bet against code in my 25-year career, and they have always lost,” says Joe Duffy, CEO of Pulumi. “AI will automate and augment coding, not replace it, thereby raising the level of abstraction that we humans operate at, considerably accelerating productivity and output.”

That’s one viewpoint. To consider others, I went back to the classics.

In Frederick Brooks’ classic book on software development, The Mythical Man-Month, he shares a study on software development productivity showing “the ratios between best and worst performances averaged about 10:1 on productivity measurements and an amazing 5:1 on program speed and space measurements.” In the 20th anniversary edition of the book published in 1995, he republishes the 1986 article, “No Silver Bullet: Essence and Accidents of Software Engineering,” which predicted that “a decade would not see any programming technique that would by itself bring an order-of-magnitude improvement in software productivity.

We don’t know yet whether copilots and other generative AI coding capabilities will exceed these benchmarks.

“The software delivery life cycle is getting disrupted by generative AI,” says Ashish Kakran, principal of Thomvest Ventures. “Dev and devops teams will become more productive with a higher percentage of team members potentially achieving outputs similar to those of 10x engineers.”

That 10x productivity gain, along with the democratization of the software development skill set, may be possible as generative AI capabilities mature and developers realign their responsibilities. “Copilots in their current form are really about developer productivity and removing that busy work,” says Ed Thompson, CTO of Matillion. “Those who assume that copilots have already fundamentally changed the job are working on the incorrect assumption that a developer’s job is to write code—it’s to solve problems.”

10 ways generative AI will transform software development

How will generative AI transform software development over the next decade? Here are 10 predictions:

1. Generating code from natural language prompts will become standard practice.
2. Code validation will become a critical developer responsibility.
3. Manufacturing will become the new development paradigm.
4. Less manual coding, but greater risks from the software supply chain.
5. New paradigms will accelerate application integration.
6. Developers will manage AI agents.
7. AI will touch multiple phases of the software development life cycle.
8. GenAI and human development personas will emerge.
9. AI will deliver operations improvements in the development process.
10. Organizations will evolve to protect themselves from AI risks.

Generating code from natural language prompts will become standard practice.

Kaxil Naik, director of Airflow engineering at Astronomer, says, “Coding will become more efficient with AI-generated boilerplate code and AI-assisted copilots translating natural language into functional code, simplifying the understanding of complex code bases and ensuring adherence to best practices.”

Stack Overflow’s 2023 developer survey shows that 70% of developers are using or are planning to use AI tools in their development process. Of those already using AI in development, over 82% use it to write code. These numbers suggest a paradigm shift in how developers will develop code, reuse existing code, and build components.

Code validation will become a critical developer responsibility.

The ability to prompt for code adds risks if the code generated has security issues, has defects, or introduces performance issues. The hope is that if coding is easier and faster, developers will have more time, better tools, and increased responsibility for validating the code before it gets embedded in applications. But will that happen?

“As developers adopt AI for productivity benefits, there’s a required responsibility to gut-check what it produces,” says Peter McKee, head of developer relations at Sonar. “Clean as you code ensures that by performing checks and continuous monitoring during the delivery process, developers can spend more time on new tasks rather than remediating bugs in human-created or AI-generated code.”

CIOs and CISOs will expect developers to perform more code validation, especially if AI-generated code introduces significant vulnerabilities. “If developers don’t implement automation to scan and monitor AI-generated code, it means exponentially more code to fix and more tech debt,” adds McKee.

Manufacturing will become the new development paradigm.

One question about using generative AI tools to write code is how it will impact tools and standards at large organizations where many development teams support thousands of applications. What will development look like in larger organizations if developers write less code but integrate more AI-generated code?

“The tooling mix across teams results in a lack of standards and complex onboarding, not to mention that it adds to the cognitive load of developers,” says Markus Eisele, leader of developer tools strategy and evangelism at Red Hat. “A blend of best practices combined with easy access through centralized developer portals is here to change this. Topped with the enriched capabilities of an application platform, this has the potential to remove friction and help with applying best practices across team boundaries.”

The implication is that integrated development environments (IDEs) may morph into assembly platforms similar to computer-aided design (CAD) in manufacturing or building information modeling (BIM) in construction. The focus shifts from building custom components to assembling preexisting ones and leveraging built-in tools to validate the design. 

Less manual coding, but greater risks from the software supply chain.

Another implication of code developed with generative AI concerns how enterprise leaders will monitor code from the software supply chain and develop policies for determining what code is permitted to include in enterprise applications. Until now, organizations have been concerned primarily with tracking open source and commercial software components, but generative AI adds a new dimension.

“Devops practitioners will play a major role in maintaining and managing the AI supply chain: the security, authenticity, and origins of AI-based models will come under more scrutiny in an enterprise’s day-to-day operations,” says Ilkka Turunen, field CTO of Sonatype. “Implementing a strategy that evaluates AI risk and properly manages an AI model’s bill of materials will help ensure proper AI hygiene and management across the devops infrastructure of any organization.”

Expect static application security testing (SAST), dynamic application security testing (DAST), and other security and code management tools to increase the automation of code scanning and help validate whether AI-generated code meets policies before developers integrate that code into enterprise repositories.

New paradigms will accelerate application integration.

Developers can expect new capabilities in integrations, which have already seen orders of magnitude of improved capabilities over the last decade through APIs, SaaS integration platforms like IFTTT, integration platforms as a service (iPaaS), and other ecosystem technologies. That said, developers still perform much work to map data fields, code transformation logic, ensure reliability, and adjust for performance considerations.

Emmanuel Cassimatis, AI and innovation team lead at SAP, says, “When it comes to integration, the development life cycle has historically been quite fragmented across its different steps, from design, build, test, integrate, deploy, deliver, and review. AI can allow unification by tapping a picture from data from different applications, resulting in greater collaboration between developers.” 

It’s only a matter of time before developers use generative AI to build “codeless,” self-healing integrations using natural language instructions and auto-generated visual flows.

Developers will manage AI agents.

Phillip Carter, principal product manager at Honeycomb, believes that generative AI will transform the tasks developers and quality assurance engineers will do in the future. “In the potentially far future, natural language is likely to guide more code generation and tests that verify generated code. If we see another massive shift in AI capabilities like the transformer, we can expect AI agents to do most of this work, with developers programming goals and constraints for these agents to follow.”

Carter continues with a bold prediction, saying, “With a new transformation that puts AI at the helm, it’s possible that programmed agents could be enabled to analyze runtime behavior for QA, observability, and security tasks to check known unknowns, something developers are often bogged down by.”

I find this prediction interesting, as it implies developers and engineers will move up the stack to define architecture, non-functional requirements, and operational requirements—guiding generative AI on developing and testing rather than writing code and automating tests.

Carter doesn’t believe in a developer-less future, though. “Humans would remain in the loop at all times, concerned more with goals, constraints, and analyzing unique circumstances,” he says.

AI will touch multiple phases of the software development life cycle.

While copilots and many generative AI tools today focus on coding, expect new capabilities to transform other phases and responsibilities in the SDLC. Humberto Moreira, principal solutions engineer at Gigster, says, “As best practices evolve for incorporating genAI into the SDLC, different models might work best for particular phases of the cycle, for example, one model optimized for requirements, one for code development, and one for QA.”

The generative AI paradigm shift is already impacting QA as tools enable more robust test cases and faster feedback on code changes.

“With the rise of AI, I think a less discussed aspect is how all the facilities around coding will witness a sea change,” says Gilad Shriki, co-founder of Descope. “It’s a matter of time before SDKs, testing, and documentation are AI-generated or assisted, which means developers will need to code and document their work in specific AI-consumable formats.”

Shriki’s last prediction suggests that developers may have to adjust their language, similar to how people must learn to speak the language that virtual assistants are programmed to comprehend. I hope this prediction doesn’t become a reality because it could mean that genAI only delivers conveniences and not necessarily productivity or quality improvements.

GenAI and human development personas will emerge.

Generative AI’s role in software development could splinter off from the roles and responsibilities currently held by human developers. Code generators, compilers, and other development tools would serve both human and machine personas.

“What’s interesting is that I think there might end up being a new view of code: one view is that traditional human view of code, the one that developers are trained and accustomed to reading and writing, but then there’s a second, somewhat hidden view, which is the AI-security-optimized, defensive view,” says Dustin Kirkland, VP of engineering at Chainguard. “This view is less readable by humans but perfectly readable by compilers and interpreters, and in this way, it becomes just another intermediate format for code.”

The question is whether alternative views will improve the ability of machine learning models to identify defects, security issues, and other problems more accurately and efficiently.

AI will deliver operations improvements in the development process.

Cody De Arkland, director of developer experience at LaunchDarkly, suggests use cases when generative AI could help improve application reliability and operations. “We can see the early signs of how dev tooling will learn from interactions, and the key will be intuitive assistance.”

De Arkland suggests that genAI could:

• Develop web application components that match the learned design standards.
• Create the feature flag as it detects a developer building a new feature.
• Stage a new software deployment (CI/CD), or roll the deployment back when it learns of problems.
• Enable real-time feedback loops to QA from customized runs instead of post-deployment.

These ideas raise the question of what next-gen devops and site reliability engineering (SRE) capabilities will be enabled or augmented by generative AI. 

Organizations will evolve to protect themselves from AI risks.

One last prediction concerns the risks of exposing generative AI to the organization’s intellectual property, including code and data. As genAI enables new capabilities across the full SDLC, it will raise new questions about whether the benefits outweigh the risks.

“As we work toward the vision of an end-to-end AI-enabled software development process, technology professionals across the board want to ensure that any code generated is of the highest quality and does not hurt the overall reliability or maintainability of the application,” says Brandon Jung, VP of ecosystem and business development at Tabnine. “With a keen eye on the data going into the model—both yours and the training set—take the time and effort to evaluate, select, and deploy AI in ways that protect your policies and your most valuable assets—your code and your data.”

The question is whether generative AI algorithms and the tools that enable them will build safeguards to protect the enterprise’s assets and how much these protections will also rely on genAI capabilities.

Although we’re still early in the generative AI era of software development, it’s becoming clear that code generation and copilots are just the beginning of new AI-enabled ways to develop, test, deploy, and maintain software.