GenAI is Key to Faster Software Development… If You Can Get People to Adopt It

There’s a lot of excited chatter about GenAI’s groundbreaking potential in software development. As with other applications for GenAI, there’s a feverish rush to get on board. For all this enthusiasm, using GenAI to accelerate software development remains a nebulous concept—hard to define and even harder to implement. We know this from firsthand experience. We’re currently using GitHub Copilot, ChatGPT, and our very own accelerator, Exadel Lumea. Each of these tools is powerful in its own right, yet with distinct advantages and limitations.

In this article, we’ll walk you through our journey of adopting GenAI for SDLC so you can see what’s possible and be inspired.

Exadel’s Journey to GenAI Adoption

Facing Up to the Challenge

Technology leaders are excited about GenAI’s potential to transform software development, but many soon face a reality check. There can be a sizable gap between their expectations and what their teams are capable of getting out of GenAI assistants. A common reason is that their teams aren’t fully embracing GenAI adoption.

Here at Exadel, we faced a similar issue. The transition to GenAI meant our developers had to change their habits. Tasks once handled manually—such as code generation, debugging, refactoring, and even drafting and analyzing requirements—were now completed more efficiently with the help of AI assistants. Automating these small but essential tasks required our developers to learn how to prompt these tools and evaluate their output effectively.

However, adoption varied across our teams. There were the innovators and early adopters eager to use GenAI, the early majority who quickly followed, and then the late majority and laggards who were slower to engage. How we addressed these different stages of adoption would be crucial to our success.

We also faced a more complex challenge than most enterprises in adopting GenAI for SDLC. Since we provide engineering services, we frequently work in the customer’s environment. This means using their toolsets and processes and following their governance and compliance procedures and policies.

This makes the overall journey more complex by an order of magnitude!

Driving Home the Need for Change

Our tech teams comprise nearly 2000 experienced professionals globally. In such a large complement, there are always early adopters of new technology. We identified those with the highest authority among their peers and involved them in a regular awareness campaign:

• We scheduled biweekly company-wide updates on our GenAI initiatives and their progress. The focus was on our GenAI adoption success stories from specific customer or internal teams.
• We tasked our various communities (Product Management, Java Engineering, Web Development, DevOps, etc.) with focusing their meetups on GenAI whenever possible.

Our community fosters innovation through discussions, meetups, and shared news, encouraging our people to explore new trends. Recently, we hosted meetups on report generation with GenAI models, AI security best practices, confident coding with AI tools, and the benefits of self-hosted vs. OpenAI LLMs.

In this way, we created internal awareness and laid the groundwork for our own adoption process. Our people received the message that this was important for the company, and with the technology’s promising potential, we could back it up with many success stories.

It Was Time to Call in the Big Guns

Understanding how persuasive credible authority can be, we invited external specialists to share their experiences:

• A Staff Software Engineer from Meta was invited to one of our AI Community meetups to discuss how they use GenAI.
• An Architect from Microsoft was invited to talk about GitHub CoPilot best practices, the most efficient way to use it, user reviews, and feedback from the broader development community.

This step was essential. We needed to prove that Exadel was not the exception and that many other promising GenAI adoption journeys had already begun across the tech industry.

Not Everyone’s a Believer

People Are Set in Their Ways

It wasn’t all smooth sailing. We found that once people made up their minds, it’s not easy to persuade them otherwise. And unless your team is well-prepared for AI adoption, any productivity improvements will be hard to achieve. We experienced this first-hand as some developers, especially those with more experience, were skeptical or even resistant. Many had tried the tools with limited success and were hesitant to give them another go. It was clear that some groundwork was needed to change these perceptions.

Once Burned, Twice Shy

There were also cases where developers who initially tried GenAI tools quickly gave up after encountering a few setbacks. Perhaps this was due to a lack of good use cases to demonstrate clear improvements in speed or quality. Regardless, they were clearly reluctant to use GenAI in other scenarios. Other team members had been willing to use GenAI but failed, not knowing how to use it correctly. They engaged the AI assistant using simple prompts but gave up after only a few attempts. This residual bias was an obvious challenge we had to address.

To counter this, we developed prompt books: step-by-step guides with use cases and templates tailored to different roles in the software development life cycle. These books aimed to give users a positive start, showing them how to effectively interact with the tools from the get-go.

Getting the Right Prompts

Now, it was time to let our people try out the tools. We picked the 20+ most common use cases for every SDLC role and added them to the above-mentioned prompt book.

For example, a standard use case for a product owner entails a user story that follows a specific format (e.g., has acceptance criteria and corresponds to the definition of ready).

For a developer, an equally relevant and widespread use case would be creating unit tests to cover a specific aspect of functionality.

It is rather tricky to produce a well-performing prompt with your first attempt if you don’t know what a good prompt looks like. Our aim with these prompt books was three-fold:

GenAI quickly proved itself useful for providing quick, context-specific answers. It added efficiency by freeing up people to do other things by handling routine tasks like writing unit tests or generating documentation. Once developers saw the benefits of AI in automating routine tasks, their initial reluctance melted away.

It Didn’t End Here

Our People Needed Training

Once we had established a good awareness level and sufficient interest, it was time to provide comprehensive training. We needed to share the full GenAI story and explain how to apply it in SDLC. We researched existing external GenAI trainings and then created our own, which are now stored in our internal Learning Management System.

We Measured the Results

We recommend defining a sufficient number of engineering best practices and agile methodologies to measure your results continuously. This gives you a sound foundation for monitoring your progress with technology adoption and its impact on your operations. It is important to measure end-to-end metrics such as throughput and time-to-market, as well as more specific metrics like bugs, accept rates, code smells, unit tests coverage, lines of code, and code vulnerabilities.

To gauge our progress, we tracked metrics such as GitHub Copilot usage and compared our acceptance rates to industry standards. We also relied on anonymous surveys to capture our people’s sentiment towards these tools and the value they added. Within six months of the launch, this approach led to all three tools being used 100% on a daily basis by our team members.

What’s Next?

We understand that the tools don’t replace human expertise but enhance it so that people can be even more productive. The positive impact of AI tools has become undeniable. Looking ahead, we know that the landscape of AI tools will continue to change. It is not a one-and-done deal; it’s a journey we’re ready to repeat as the technology evolves.

Along the way, we identified gaps in the current AI toolsets, which prompted us to develop our own GenAI-based solution to accelerate the SDLC. The GenAI toolset landscape will likely remain fluid over the next few years, so we will have to adapt our work habits as a new generation of tools becomes available.

Our experience has also shown us that ‘evangelizing’ and encouraging adoption is not always enough. Sometimes, clear expectations must be set, and people need to be told directly that these tools are essential. A blend of inspiration and firm expectation has been key to our adoption strategy. A fundamental shift in thinking is required.

We now understand that the steps in this GenAI adoption journey will have to be repeated multiple times in the foreseeable future. However, we’re not daunted, as we now have a sound strategy to deal with it. We hope our experience will inspire you to develop your very own GenAI adoption strategy!