Domain Expertise is an Engineer’s Knight in Shining Armour

The conversation surrounding AI and job displacement often gets reduced to a neat equation: smarter AI equals fewer human jobs. It’s a narrative that assumes AI functions like an end-to-end solution, where requirements are fed in and finished products are received. 

However, several experts in the industry today argue otherwise. Among them is Balaji Srinivasan, president of the Network School and a venture capitalist, who stated in an X post, “AI doesn’t do it end-to-end. It does it middle-to-middle. The new bottlenecks are prompting and verifying.” 

Srinivasan believes the imbalance lies in how these tasks scale. “AI prompting scales, because prompting is just typing. But AI verifying doesn’t scale, because verifying AI output involves much more than just typing,” he added. 

In his view, the current AI ecosystem is excellent for front-end work, where visual aspects can be easily checked with the eye. But for anything more intricate, like code or text, accurate evaluation requires deep subject knowledge. 

“That means knowing the topic well enough to correct the AI,” he said. 

Srinivasan also stated that the concept of verification for AI, especially for end users, remains underdiscussed. 

“Yes, you can try formal verification or critic models where one AI checks another, or other techniques. But to even be aware of the issue as a first-class problem is half the battle,” he added. His thesis has found resonance across the industry. Among those agreeing was David Sacks, appointed as the ‘AI czar’ by the Donald Trump administration in the United States. 

For a more nuanced perspective, AIM reached out to Jonathan Bryce, executive director at the Cloud Native Computing Foundation (CNCF), which maintains open-source frameworks like Kubernetes. 

Drawing from his three decades of experience in software development, Bryce believes the shift AI brings is similar to those the industry has seen—moving from low-level languages to high-level languages,  rather than a total replacement of jobs. 

“If you go back that far, to write code, you had to write at a much lower level, which was slower, more error-prone, harder to test, harder to manage the performance, and something that just fewer people had access to,” Bryce said. 

“As we moved up through higher-level languages, like C and C++, and then for the web, PHP and Python and Rails and these kinds of things, rather than there being fewer developers because it was easier to write software, and we have many, many more developers,” he added. 

Like Srinivasan, Bryce said that understanding how to design a system and verify it will become a new skill set that is more critical than knowing the exact syntax for defining a variable, scoping it or formatting a function.

What also supports the above statements is the amount of research being conducted concerning hallucinations in AI systems. 

A quick look inside the arXiv database of technical papers shows a growing number of research studies on hallucinations in AI systems. Furthermore, one of the most publicised aspects of OpenAI’s recent GPT-5 release was that there were “significant advances” in reducing hallucinations. And yet, this doesn’t involve eliminating them entirely, and leading AI models today continue to hallucinate. 

Agentic AI Might Add Newer Jobs

AIM also reached out to Mukund Kalmanker, global head of data analytics and AI practice, at Apexon, a US-based tech services firm. 

“A key challenge is what we call the ‘context catastrophe’. AI can execute large volumes of tasks and perform logical reasoning, but it often misses the situational understanding that human experts intuitively bring,” he said. This is also a reason why concepts like “context engineering” have been on the rise since the last few months. 

He further stated that this becomes clear in edge cases where AI fails simply because the parameters were not explicitly defined in the prompt. 

“This new paradigm also calls for reimagining roles. We are seeing the emergence of new functions such as AI agent supervisors, who manage performance across multiple agents; context engineers, who translate business goals into effective prompt logic; verification specialists, responsible for ensuring AI systems align with strategic and ethical standards; and collaboration specialists, who orchestrate smooth handoffs between human and automated processes,” Kalmanker said, pointing to a popular study from the World Economic Forum (WEF), which presents a more optimistic view where AI can eliminate 92 million jobs, but can also lead to the creation of 170 million new ones. 

Having said that, it is also important to view the WEF’s findings with a fair perspective, which also emphasises how certain software development jobs, customer support and finance roles involving repetitive tasks are being adversely affected. 

Furthermore, several developers have noted how tools like Claude Code are automating various tasks independently, and even CEOs like Marc Benioff from Salesforce suggest that AI is managing 85% of the company’s customer service interactions. Several reports have also highlighted how entry-level roles are being submerged due to the advent of AI. 

However, when it comes to more nuanced and end-to-end tasks, humans with deep domain expertise remain essential to verify and validate—until AI systems are fine-tuned to shoulder that responsibility. 

The post Domain Expertise is an Engineer’s Knight in Shining Armour  appeared first on Analytics India Magazine.

Generated by RSStT. The copyright belongs to the original author.

Source