Gen AI has entered the product engineering lifecycle

From ideation through launch, generative AI can introduce levels of productivity to product engineering that product businesses simply cannot ignore.

Product engineering has seen dramatic change in the last few years, from agile methodologies, DevOps and design thinking in software development, to modular design principles, digital twins and 3-D printing in hardware.

The result has been a product engineering renaissance that’s given rise to newly invigorated and turbo-charged product companies.

But even more change is ahead for hardware and software businesses with the arrival of generative AI. From ideation through launch, generative AI can introduce levels of productivity to product engineering that product businesses simply cannot ignore. 

Using generative AI, for instance, product designers can explore ideas beyond the realm of their own imaginations and experiences, then develop initial design concepts significantly faster than with traditional methods.

This applies to both new products and adaptations of existing ones. For example, if a business needs to modify an existing product for a particular customer segment, product engineers could ask a gen AI agent to analyze the existing product and suggest alternatives. As the product modifications took shape, the gen AI agent could continue to refine the product, check it against regulatory requirements, generate blueprints and prototypes, and keep other team members in the loop, including sales and marketing.

Gen AI in the product engineering software development lifecycle

As for software development, it’s not a question of whether product businesses should embrace generative AI—but how best to do it.

From requirements analysis and systems design and coding, to testing, to deployment and maintenance, we’re seeing productivity gains of 30% to 40% in the software development lifecycle with the use of generative AI. We expect that in the next few years, those gains will spike even higher.

The most common areas in which we’re seeing businesses apply generative AI to software development include:

• Automated code generation: Because AI can learn to replicate the syntax, patterns, practices and nuances of human code in various programming languages, it can convert conversational prompts into full or partial lines of functional code. Developers are then able to accept or reject these code suggestions.
  
  AI auto-completion tools can speed development even for experienced developers who would otherwise have to spend time typing the code. It can also be useful for developers when they’re using a language whose syntax they’re not completely familiar with.
  
  Using natural language prompts, developers can also ask a gen AI chatbot to explain code, improve syntax, provide ideas, generate tests, and modify existing code.
  
  
• Automated documentation: With time-to-market pressure, documentation often falls by the wayside. As a result, code libraries are often left undocumented, making them difficult to comprehend.
  
  But because of gen AI’s ability to summarize and generate text, it can create new documentation, even for code written decades ago by people who are no longer at the company. It can also update documentation when new features are added and translate documentation into multiple languages.
  
  Not only does this speed documentation, but it also helps business avoid the scenario of having to restart projects because they don’t know how an existing but undocumented component works.

• Testing. Code testing has become less manual with scripted and data-driven testing, but there is still a level of manual effort involved. Generative AI, however, automates the creation of highly effective test cases for an array of scenarios.
  
  AI can also generate synthetic data for cases involving sensitive data. As the term implies, this data is created digitally rather than being gathered from real-world events. While synthetic data contains none of the original data from which it was derived, it retains the qualities of the original data; from a statistical standpoint, anything you do with it— such as building a predictive model—will produce the same results as if the original data were used.
  
  In healthcare, for example, synthetic data wouldn’t contain any actual patient data, so privacy regulations like HIPAA or GDPR would not apply to it.
  
  By using synthetic data, companies can address privacy concerns, overcome data scarcity, and accelerate the development of AI applications across various industries.
  
  
• Code reviews and debugging assistance: A rule of thumb is that developers spend 25% to 50% of their time debugging code, making this a prime area for productivity improvement. Gen AI tools can be trained on large numbers of software bugs and bug fixes, enabling them to analyze code, flag problems and errors, and offer suggestions for resolving issues.
  
  This capability saves time and improves code quality, reliability and security, avoiding errors before they cause problems.
  
  Gen AI can also conduct code reviews, proactively scanning changes to identify risks that human reviewers may miss and flagging areas for improvement. Code reviews are an integral part of producing high-quality software. Potential improvements range from optimizing code to be more efficient or consume less memory, to detecting issues that could lead to a security breach.
  
  
• Refactoring recommendations: Poor-quality code is a major cause of high technical debt. Because gen AI can learn code logic and context, it can also improve code readability and make it less complex. It can point out inefficient algorithms and other performance bottlenecks and can suggest alternatives.
  
  The resulting code is of higher quality, which lowers technical debt. Moreover, once the code is refactored, developers  will grasp its inner workings much more quickly.
  
  One caveat: the trick in refactoring is doing so without impacting the code’s behavior or functionality. Developers would need to check AI outputs to ensure this is not the case.

We worked with a healthcare client, for instance, with highly complex health plan requirements. Traditionally, the company would need to print out the requirements document and manually highlight all the business rules that needed to be extracted, coded, tested and deployed. 

Now, a generative AI program automatically extracts the business rules, creates a configuration file, applies it to the system and tests it. Software developers are still involved, but the job is far less manual, resulting in dramatic acceleration and productivity gains. 

Caveats of gen AI in software development

It’s important to note that realizing generative AI productivity gains in software development requires more than just giving a developer a gen AI assistant or copilot. Developers should also be armed with reusable templates and libraries of leading frameworks and models to streamline workflows and ensure consistency. Automated security reviews and compliance checks can further ensure adherence to standards, and code explainability can improve resilience and reliability.

Product engineers should also be thoroughly trained in how and when to use these tools, as well as how and when to check gen AI outputs. For instance, it may be tempting to provide junior-level staff with these tools, as they may jump to higher levels of productivity than more senior developers. This was apparent in a recent Stanford and MIT study in which call-center reps who used gen AI were 14% more productive on average than those who didn’t. The gains were even greater among workers who had been on the job for less than a few months. However, junior-level staff will also need more assistance in vetting what the tools produce. 

Generative AI in software development will also require a fundamental culture change. Traditional coding is very deterministic: you write a set of instructions, and those instructions produce an output. 

With gen AI, the process is much more organic. As inputs change and the models continue to learn, the outputs need to be checked and validated on an ongoing basis. This requires a whole new way of thinking and a new approach to building out the software engineering organization.

Finally, businesses must ensure the code the AI generates is secure and free of copyright issues. If the code base the AI was trained on had a security issue, the code that's generated will potentially have it too.

Businesses can avoid such scenarios by understanding where their code comes from, what it was trained on, the models that were used, and where it’s deployed. And if a problem is detected, they need the ability to stop it in its tracks. 

Productivity gains of generative AI in product engineering

The immediate productivity benefits of generative AI in product engineering will collectively contribute to a more streamlined and rapid development process in which teams can build, maintain and update applications at an accelerated pace. 

But the end goal isn’t necessarily cost savings or labor reductions. Instead, it’s the ability to do more in the same amount of time. This is essential for product businesses—getting features and functions and products into the market more quickly. In a McKinsey study, generative AI accelerated product time-to-market by 5%, improved product managers’ productivity by 40% and boosted employee experience by 100%.

Further, the use of gen AI in product engineering will improve the quality of life for software developers, freeing them from mundane tasks to focus on more complex and engaging ones—such as problem solving and working with generative AI to get it done.

In the end, it will be less a matter of competitive advantage for product businesses that best use generative AI than a vast disadvantage for those that don’t.

To learn more about how product engineering organizations can take advantage of generative AI, tune into our product innovation podcast series.