When the federal government launched its Health Insurance Marketplace in 2013 as part of the Affordable Care Act, the rollout was plagued with IT issues. The website — HealthCare.gov — received millions of visits in a matter of hours, causing a swift and conspicuous crash. And that was only the beginning. Exacerbating the site’s failure were an incomplete design, a buggy user experience, and disastrous data transfers that created chaos and confusion for insurers.
To fix the website, the Centers for Medicare & Medicaid Services (CMS) hired Accenture Federal Services in January 2014.
“Within eight weeks, Accenture delivered significant technical improvements to the website, stabilizing it during the peak of [its] initial enrollment period,” Accenture said of its work in a 2020 case study.
“This enabled millions of Americans to enroll in health insurance, many for the first time.”
The reason Accenture succeeded where the federal government failed boils down to a single, vital IT discipline, according to software strategist Kolapo Akande: performance engineering.
“Performance engineering is the way you think about the performance journey by looking at the design, tests, implementing and deployment of any application, software or hardware,” said Akande, who worked on the Health Insurance Marketplace when he was a performance architect at Accenture.
“It’s everything from capacity planning to testing your software development lifecycle process,” he said.
Akande is founder and CEO of Pledge Software, a builder of a web-based application that connects donors to nonprofits. He said he sees ample opportunity for organizations today to use the same performance-engineering techniques that Accenture used.
In fact, performance engineering may be even more important now than it was then on account of artificial intelligence, suggested Akande, who said AI brings additional concerns that demand the unique talents of a performance engineer — things like CPUs, GPUs, nonlinear scaling of load and more distributed testing/architecture.
But the rapid rise of AI isn’t merely shining a fresh spotlight on IT performance engineers. It’s also demanding that they grow and evolve, driving many who work with hybrid cloud data centers to harden their skills and learn new methodologies for optimizing data systems, according to industry experts.
AI’s Impact on Performance Engineering
Steven Keith Platt, site director for the Institute for Data, Econometrics, Algorithms and Learning (IDEAL) said AI is becoming part of the base layer of software engineering. Because emerging AI-first infrastructure functions like an operating system — for example, it coordinates GPUs, schedules training and inference, and allocates data and compute at scale — the software design process is being reshaped, he noted, emphasizing impacts for performance engineering with regards to both methods and challenges.
“The transition [to AI] exposes hard constraints, including power density, interconnect bandwidth and cost/latency, which is pushing advances in specialized hardware, scheduling and model serving,” said Platt, who is also a professor at Loyola University Chicago’s Quinlan School of Business.
“This does not replace Windows or Linux. It is an AI-optimized layer that sits alongside and atop platforms. The architecture is still early. Evaluation, observability, data governance and reliability remain uneven, and organizations are working through integration with existing systems.”
The software-defined infrastructure company Liqid has determined that performance engineering is as useful for designing IT infrastructure as it is for designing software, according to Liqid President and Chief Strategy Officer Sumit Puri, who cited the need for high-power resources in data centers due to the increasing use of AI. He said the company focuses on centralizing GPU pools to help with performance engineering; by using compostable approaches, it maximizes GPUs to improve performance while also reducing waste, he noted.
“I need that server to have a lot of GPUs because I need it to run very fast,” Puri said. “We’re not limited by two or four or eight in a box. We can compose 30 GPUs to a server and give you the fastest servers on the planet.”
Upskilling for Performance Engineers
To get the Health Insurance Marketplace running smoothly back in 2014, Accenture indicated that it had to stop the site’s bleeding.
“To stabilize the site, Accenture completely took over maintenance and operations … The team tackled defects and delivered urgently needed fixes,” it reported in its aforementioned case study.
“To keep the system fully functioning, especially during peak enrollment, Accenture provided continual monitoring and reporting for HealthCare.gov. Each week, the team delivered multiple software releases, with significant enhancements.”
To make enterprise AI successful, performance engineers today need to focus less on fixing platforms that are slow or inefficient and more on building platforms that are designed for high performance from the start, suggested Akande, who said doing so might require practitioners and employers to invest in upskilling.
A good place for performance engineers to start, Akande noted, is developing a soup-to-nuts understanding of the architecture, software and hardware that’s needed to fuel AI-based systems.
“During performance tests, you need to understand the different parts of the systems [and] know where to start, such as the CPU,” Akande explained.
“If you understand the volume of data and where it is stored, then you know where to focus your testing effort.”
As the development and use of AI-based workloads accelerate, Akande expects that performance engineers will need to fully understand the technology behind distributed architectures. Engineers who are skilled in Kubernetes, pre-Kubernetes deployment, CEI, continuous integration and continuous deployment are more likely to stay ahead of the curve, he suggested.
Performance Engineers: Leading Transformation
According to IT leaders like Akande, Platt and Puri, performance engineering continues to evolve from fixing performance problems after the fact to designing systems for high performance from day one. As a result, they observed, engineers will likely evolve into architects of efficiency, scalability and resilience. By using new skills, tools and approaches, they indicated, performance engineers can lead their organizations through transformation while ensuring that AI-driven applications perform at the highest possible level.
And yet, performance engineers acknowledge that they can’t succeed alone. What they needed a decade ago in the context of healthcare reform — “one cohesive team that worked collaboratively and transparently,” Accenture stated — is what they also need today in the context of AI, data suggests.
A recent study in the journal Frontiers of Computer Science, for example, found that “collaboration intensity is a critical determinant of performance gains” in software development teams.
“Everyone was open about schedules, risks and defects, and all worked together to share knowledge and solve problems proactively,” stated Accenture, explaining its work with CMS, of which performance engineering was a critical part.
“Instead of finger-pointing, people were rewarded during daily meetings for candidly acknowledging, ‘This isn’t working…and here’s what we need to do to fix it.’”
Jennifer Gregory writes about B2B technology for clients including Microsoft, IBM, Salesforce, Verizon, Google and AT&T. She lives in Raleigh, North Carolina, and rescues homeless dachshunds in her spare time.
© 2025 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.