AI is often discussed in the language of replacement, engineers will write less code, recruiters will screen fewer resumes, agents will perform tasks that once required teams of people, but inside LinkedIn, the more consequential change may be happening somewhere less visible. AI is beginning to alter what the company considers valuable human work.
Nithya Rajagopalan, Head of LTS Engineering at LinkedIn India, describes the transition as the latest stage in an evolution she has witnessed over years of engineering. On premise infrastructure gave way to cloud, and now engineering is moving towards an AI native model.
Yet her account also exposes an uncomfortable contradiction in the current AI narrative. If machines increasingly handle execution, the responsibility for deciding what should be built, whether it is correct and whether it is safe does not disappear. It moves upwards. “Having very strong foundations is very much important,” Rajagopalan says.
For an engineering workforce entering an era of coding agents and agentic workflows, that may be more than a statement about skills. It is a warning against confusing the ability to generate software with the ability to engineer it.
AI changes the job before it changes the job title
LinkedIn’s engineers are increasingly using AI tools to accelerate software development. Rajagopalan says the organisation is becoming AI native, with engineers expected to use AI to make their work faster while improving quality. But speed is only one part of the change.
The engineer is increasingly expected to understand the customer problem and determine how it should be solved. “What is the problem customer needs and how can we solve it?” she asks.
That shift matters because AI can potentially compress the execution layer of software development without eliminating the need for judgement. If anything, the opposite may be true. When machines can produce multiple possible solutions quickly, the scarce skill becomes knowing which solution is worth pursuing.
Rajagopalan says product thinking is consequently becoming more widespread across engineering. Engineers are being expected to understand the customer before applying their technical knowledge and AI tools to the problem.
The implication is significant. AI may reduce some of the traditional boundaries between engineering and product teams, while simultaneously making engineering expertise more demanding.
The agentic workforce arrives first inside engineering
The next stage is already visible in LinkedIn’s internal workflows. Rajagopalan points to coding agents and other AI enabled tools being used across engineering. The objective is not restricted to generating code. LinkedIn has a large software environment made up of multiple products and microservices, and AI driven workflows are being used to help engineers move across these systems while addressing larger customer problems.
“We are adopting that so that we can develop with better quality and faster,” she says. That formulation is revealing. The promise of agentic AI is frequently framed around autonomy, but LinkedIn’s engineering use case, as described by Rajagopalan, remains centred on augmentation.
The agent does not become the engineer. It becomes part of the engineer’s operating environment. That also creates a new responsibility. Rajagopalan adds that engineers now need to understand the nuances of regulation and compliance and how those considerations should be incorporated into products.
In other words, AI literacy is no longer just about knowing which tool to use. It increasingly involves understanding what the tool is allowed to do and how its output should be governed.
The hiring assistant exposes the harder AI question
The same tension appears in hiring. LinkedIn already has AI capabilities for candidates, including tools intended to assist with interviews and help build profiles. On the recruiter side, its hiring assistant draws on information from LinkedIn’s Economic Graph, including skills, experience and education, along with applicant information available through customer ecosystems and applicant tracking systems.
The productivity proposition is straightforward. Rajagopalan says recruiters can potentially examine substantially fewer profiles before identifying relevant candidates. But this is where the distinction between automation and judgement becomes critical.
“Now you spend more hours doing that to do the human side of the work rather than the mundane work,” she says. The intended outcome, therefore, is not the removal of the recruiter. It is the redistribution of the recruiter’s time.
That sounds straightforward, but it places greater importance on the quality of the machine’s filtering. If an AI system reduces the number of profiles reaching a human, the quality of what gets filtered out becomes just as consequential as the quality of what gets surfaced.
Rajagopalan says LinkedIn conducts intensive evaluation before AI outputs are surfaced or used in decisions. She also stresses privacy and security and says information is not shared between customers because of LinkedIn’s tenancy model. “We have a very, very strong tenancy model,” she points out.
The productivity question remains unresolved
This is also where the broader AI productivity argument becomes less straightforward. Rajagopalan does not frame engineering impact around token consumption or the amount of AI generated output. Her benchmark is whether the customer problem was solved. That is arguably a more meaningful test than raw usage metrics. But it also makes the absence of detailed before and after measurements notable.
Rajagopalan does not provide specific engineering productivity figures in the interaction. She refers instead to examples of workflows where AI has helped reduce the time required for particular tasks.
The distinction matters because faster does not automatically mean better. Generating code more quickly does not establish that a product is better engineered. Screening fewer candidates does not establish that the eventual hire is better. And deploying agents across workflows does not by itself demonstrate that an organisation has become more productive. The real question is what organisations do with the capacity AI creates.
The human role is moving upwards
Rajagopalan’s account suggests that the immediate impact of AI at LinkedIn is less about eliminating established roles and more about changing their centre of gravity.
Engineers need technical foundations, but also AI fluency. They need to understand products and customers, not merely systems. They increasingly need awareness of regulation and responsible AI. Recruiters may spend less time on repetitive screening and more on judgement. That is a considerably more complicated picture than the simple claim that AI will replace white collar work.
AI can take over portions of a workflow without taking responsibility for the outcome. That responsibility remains human. And as AI becomes more capable, the value of the human contribution may increasingly lie not in doing the work manually but in defining the problem, challenging the machine’s answer and deciding whether the result is good enough to ship or act upon.
The technology may be getting better at doing the work. The people using it are being asked to get better at deciding what work is worth doing in the first place.