EPAM's Adam Auerbach on AI, Cloud Agility, and Testing

What happens when AI meets software development? According to Adam Auerbach, Vice President at EPAM, the result is a perfect storm of innovation, efficiency, and growth. In an interview with Dataquest, Auerbach shares his vision for an AI-driven software development future. 

Auerbach's vision for the future of software development is one where AI agents, acting as virtual assistants, streamline complex tasks. These agents can generate code, write test cases, and even debug errors, freeing up human developers to focus on higher-level problem-solving and creative endeavors.

"AI agents are not here to replace developers," Auerbach clarifies. "Instead, they are tools that augment human capabilities, enabling teams to deliver higher-quality software at a faster pace."

When asked about his inspiration for focusing on AI, Cloud Agility, and Testing domains, Auerbach reflects on his early days at EPAM, where the initial mission was to help clients adopt DevOps and achieve continuous delivery as part of their cloud transformation. “Moving to the cloud fundamentally changes how teams work, and their adaptability plays a critical role in realizing the full benefits of the cloud,” he says.

EPAM’s approach combines engineering and consulting to evaluate clients’ workflows and key performance indicators, crafting strategies to enhance automation and efficiency. The rise of generative AI has been a pivotal moment, reinforcing EPAM’s vision of enabling high-performing engineering teams. Tools like GitHub Copilot, Auerbach explains, have demonstrated how AI can enhance productivity and streamline tasks, from writing user stories to creating test automation.

“By incorporating AI agents, organizations can standardize best practices, accelerate their DevOps maturity, and make tangible progress toward continuous delivery,” he notes. This alignment of AI with cloud agility and testing not only enhances productivity but also drives better outcomes as teams evolve.

India: EPAM’s AI-Enabled Hub

India stands as a beacon in EPAM’s AI journey. The country’s professionals have embraced AI tools and practices with unmatched enthusiasm, establishing India as the company’s AI-enabled hub.

"India's willingness to embrace change and experiment with new technologies has made it a hotbed for AI innovation," Auerbach notes. "We're seeing rapid adoption of AI-enabled processes across various industries, from finance to healthcare."

“India leads EPAM’s global AI strategy, with over 1,300 prompt engineers in the testing practice alone,” Auerbach reveals. These teams have developed proprietary products like EPAM’s EliteA™ and executed  numerous proof-of-concepts (POCs) to integrate AI into projects. From becoming the destination of choice for AI-related initiatives to advancing research and development efforts, India’s contributions have been pivotal.

EPAM’s focus on upskilling professionals and fostering an AI-first mindset has allowed its Indian workforce to take the lead in deploying innovative solutions. This strategic emphasis on India also underscores the growing prominence of the country in global technology ecosystems.

Overcoming AI Adoption Challenges

Despite AI’s potential, Gartner’s observation that 50% of AI projects fail to transition from POC to production underscores significant challenges, often stemming from unclear objectives, data quality issues, and resistance to change. Auerbach acknowledges this hurdle, emphasizing the importance of delivering tangible results.

“Many of our clients have successfully transitioned to using AI solutions in production,” he says, citing a major Canadian retailer with over 700 users actively leveraging EPAM’s EliteA™ platform. EPAM’s approach—integrating AI directly into project delivery—has accelerated adoption by delivering measurable cost savings and productivity gains.

Auerbach also highlights that AI’s faster uptake in managed services scenarios, where clients immediately benefit from efficiency improvements, contrasts with the longer lead times for engineering use cases.

Generative AI in the Software Development Lifecycle

EPAM’s exploration of generative AI agents, such as GitHub Copilot and large language models (LLMs), has opened new possibilities across the software development lifecycle (SDLC). Testing, in particular, has emerged as an ideal starting point.

“An AI agent can take user stories, generate corresponding test cases, and publish them to a test management tool. For instance, in a retail project, the AI agent could extract requirements from user stories to generate test cases, upload them to a platform like Jira, and create corresponding automation scripts for Selenium, reducing manual effort and increasing accuracy. It can then create automation code, check it into the repository, and even run and debug tests,” Auerbach explains. This approach addresses bottlenecks caused by manual testing processes, enabling teams to achieve higher productivity and reduce rework.

Industries like retail and financial services are leading the adoption of AI agents, while life sciences are cautiously exploring AI’s potential due to privacy concerns. Auerbach shares an example of a global pharmaceutical company using AI agents for internal Level 1 support, paving the way for broader applications.

The Human Element: Still Indispensable

With AI performing tasks like code generation and debugging, the role of human developers is evolving rather than diminishing.

Auerbach firmly believes that AI acts as an enabler, freeing engineers to focus on more strategic tasks.

AI tools are most effective when used in conjunction with human expertise. Developers must still be involved in the process to provide guidance, make critical decisions, and ensure the quality and security of the software.

“Since AI agents handle repetitive work, developers can spend their energy ensuring applications handle unexpected scenarios, ultimately improving product quality,” he says. However, the human aspect remains critical, as teams must adapt their workflows to fully embrace AI’s capabilities.

Risks and Limitations of AI Agents

While AI agents hold immense promise, they are not without limitations. The quality of input data heavily influences their effectiveness. Ensuring high-quality input data requires strategies such as rigorous data validation processes, cleansing inconsistent records, and training teams to create structured and meaningful documentation.

For example, a retail organization can enhance AI effectiveness by standardizing customer interaction logs, enabling the AI to derive better insights and provide accurate recommendations. Poorly structured documentation or insufficient data can limit AI agents’ output quality. Organizational culture also plays a crucial role, as resistance to change can hinder AI adoption.

As AI becomes more integrated into software development, it's crucial to address potential challenges and ethical considerations:

Data Quality and Bias: AI models rely on high-quality data to make accurate predictions. Biased or incomplete data can lead to biased and inaccurate outcomes.

Security and Privacy: AI systems must be designed and implemented with robust security measures to protect sensitive data.

Job Displacement: While AI can automate certain tasks, it also creates new opportunities for skilled professionals who can work alongside AI systems.

India’s Enthusiasm for AI

Auerbach commends Indian professionals’ willingness to embrace new technologies, which accelerates innovation. From enhancing test cases with Retrieval-Augmented Generation (RAG) to modernizing workflows, India’s rapid adoption of AI-enabled processes has positioned it as a hub for global AI innovation. EPAM’s investment in talent development and innovation in India underscores the strategic role the country plays in shaping the company’s AI journey.

Advice for CIOs

For CIOs contemplating AI integration, Auerbach’s advice is clear: start now. “This is a multi-year journey that involves understanding AI agents, implementing them, and adjusting workflows. Delaying adoption risks falling behind competitors,” he warns.

Conducting a value stream analysis and engineering excellence assessment can help CIOs identify areas where AI would provide the most value. Auerbach emphasizes the need for a prioritized approach and clear objectives to avoid wasting resources.

As AI continues to evolve, we can expect further breakthroughs in the following areas:

Specialized AI Models: Smaller, more focused AI models tailored to specific organizational needs will drive greater efficiency and accuracy.

Advanced AI Agents: AI agents will become more sophisticated, capable of handling increasingly complex tasks and making autonomous decisions.

AI-Driven Automation: Automation will permeate every aspect of the software development lifecycle, from code generation to testing and deployment.

By embracing AI, software development teams can unlock new levels of productivity, creativity, and innovation. As Auerbach concludes, "The future of software development is AI-powered, and those who adapt to this new reality will thrive."

Looking to the future, Auerbach foresees the development of smaller, specialized language models tailored to individual organizations. These models, which will be less resource-intensive than current LLMs, hold the potential to drive greater efficiency and innovation in software development.

Original article here.

Discover how EPAM helps companies become an AI-first business to thrive in the next wave of disruption: https://www.epam.com/services/artificial-intelligence