AI is fundamentally rewriting age-old codes and reshaping how systems are designed, tested, and trusted. At a massive scale, traditional engineering assumptions are breaking down, and nowhere is this more evident than in testing.
Writing unit tests has increasingly been delegated to AI tools like GitHub Copilot and ChatGPT, offering developers a much-needed sense of respite by reducing repetitive effort and speeding up test creation.
This shift is structural, and it has sweeping implications for engineering talent, infrastructure design, and how the next generation of technologists must be trained.
Venkat Pullela, CTO, networking at Keysight Technologies, mentions in a conversation with AIM, “With AI, people are forced to do system testing. You cannot do unit testing anymore.”
However, the larger question is why unit testing is no longer enough.
For decades, engineering excellence was measured by how well individual components performed in isolation. Unit testing became the gold standard. But AI systems, especially those running on thousands or even millions of GPUs, do not fail that way.
“The failures at a system level are fundamentally different,” Pullela explains. “And people are finding failures that are unique and different.”
In fact, “during interviews, candidates with an existing code block and ask them to explain its time and space complexity, evaluate trade-offs, and propose alternative approaches. Rather than testing how fast they can write code, we focus on how deeply they can reason, analyse, and think critically about it,” founder and CEO of a software technology company tells AIM.
In hyperscale AI environments, even a minor anomaly can cascade across the entire system. “When you have a million GPUs, even if one GPU runs at half speed, all the million minus one also is as if they are running at half speed because of it.”
The cost of such failures is enormous. It forces teams to rethink how early and how holistically they test.
Perhaps the most radical change is when testing now begins. Engineers are being asked to validate entire systems before hardware even exists.
“You don’t even have an ASIC (Application-Specific Integrated Circuit),” Pullela notes. “You have a design of an ASIC—and you have to do system testing.”
To make this possible, companies are blending simulation, emulation, and real components into what is often loosely called a ‘digital twin’. But the intent is precise: bringing system-level behaviour forward in time.
“We are combining simulation, emulation and real components and building a system,” he says. “You are bringing the system to you while you don’t have anything—you just have ideas.”
This shift-left approach is dramatically compressing development cycles, uncovering failures earlier, and fundamentally changing how products reach production.
This transformation is not happening in silos. Vendors, cloud giants, and infrastructure providers are now tightly coupled in co-design relationships.
Hyperscalers have become lighthouse customers, shaping architectures, testing methodologies, and tooling alongside their partners. Massive-scale simulation environments, such as containerised networks that mirror real-world deployments, are now standard practice before a single line of production code is released.
The Other Side
The Indian IT Sector’s testing and quality assurance/quality control (QA/QC) function currently has over 3.75 lakh active professionals across different experience ranges as of April 30, 2025, according to Xpheno data.
The talent pool’s churn, often seen as an indicator of active hiring activity, has been in the average range of 7-9%.
Meanwhile, a Reddit discussion thread reflects widespread scepticism within the QA community about claims that AI can fully automate end-to-end testing or replace QA roles in the near future. While AI is increasingly being adopted as a productivity booster, most practitioners see it as an assistive tool rather than a substitute for human judgement.
Several experienced QA professionals and Software Development Engineers in Test (SDETs) noted that, despite heavy marketing, no AI-powered testing tools today can independently design, execute, and maintain meaningful end-to-end tests at scale without significant human guidance.
AI performs reasonably well in limited areas such as generating helper functions, writing boilerplate test code, parsing unfamiliar codebases, creating mock data, or assisting with debugging. However, when tests involve complex workflows, business logic, domain-specific edge cases, or evolving product behaviour, AI’s effectiveness drops sharply.
Attempts to fully automate testing by simply pointing AI tools at an application often yield brittle, low-quality tests unless a skilled QA professional actively directs the process. Without human oversight, AI-generated tests are often deemed unreliable.
Skills are Lagging the Pace of Change
As systems grow more complex, the skills required to build and validate them are evolving faster than institutions can adapt.
When systems become more complex and AI-generated code becomes more common, the need for professionals who can validate behaviour, assess risk, and understand system impact is expected to grow.
Upskilling is widely encouraged, not as a way to escape QA, but to strengthen it. Many QA professionals are learning to work alongside AI by using it for acceleration while focusing their own efforts on higher-value work such as test strategy, exploratory testing, business validation, and cross-functional collaboration.
This naturally pushes QA roles closer to SDET or Dev-in-Test profiles, with stronger coding skills and AI-assisted workflows.
“The skill sets needed are evolving so fast,” Pullela admits. “None of us has the skill set. Any given time, we all feel less adequate than before.”
Even seasoned engineers are re-learning fundamentals, often in real time. “This is like a once-in-a-lifetime opportunity,” he says, explaining why many senior technologists are postponing retirement to stay in the game.
The challenge is even sharper for new graduates. Traditional curricula—heavy on theory, light on real systems—are proving insufficient for an AI-first world.
As AI systems become more interconnected, engineering is once again becoming collaborative and interdisciplinary. Those who cannot explain their thinking—or challenge others—risk becoming invisible, regardless of technical ability.
The End of Unit Testing Is Really the Beginning
The decline of unit testing as the primary quality gate does not mean testing is disappearing. It means testing is becoming more expansive, more expensive, and more critical than ever.
Engineers are now judged by how well they understand systems, anticipate failure domains, and validate behaviour at scale—often before hardware exists.
In that sense, AI is not killing engineering fundamentals. It is raising the bar. And for those willing to learn fast, think system-first, and build relentlessly, the opportunity—like the technology itself—is massive
The post AI is Forcing the End of Unit Testing. Here’s What It Means for Engineering Talent appeared first on Analytics India Magazine.
