Recruitment has always been a balance between efficiency and empathy. Technology promised to ease the process, yet it often left recruiters bogged down by repetitive tasks. The real goal, finding the right talent and building relationships, often slipped out of reach.
Agentic AI is changing that. Beyond static chatbots, digital agents are beginning to reshape hiring. Large language models (LLMs) have triggered a wave of innovation, but using them in production is far from simple.
Traditional workflows fail to capture the adaptive and memory-driven nature of agents. Enterprises now recognise that to unlock value, agents must be designed like distributed systems, striking a balance between quality, scale and latency.
At Cypher 2025 in Bengaluru, Karthik Ramgopal, tech lead of product engineering and distinguished engineer at LinkedIn, said the company’s AI-powered Hiring Assistant was more than a product launch. He stressed that the assistant was meant to bring purpose back to work, not to replace recruiters.
When LinkedIn CEO Ryan Roslansky announced it, he positioned it as a bet that AI could streamline the tactical side of recruiting while enhancing the human experience.
“We believe this brings joy back into recruiting,” Ramgopal said. “It lets people focus on the strategic aspects of their job while delegating routine tasks to the hiring assistant, all while staying in control.”
Agents as Systems
Once agents move from prototypes to production, they start to resemble distributed systems. They face challenges such as state consistency, fault tolerance, and downtime management. These are amplified by the non-determinism of LLMs. Ramgopal argued that ignoring these realities leads to fragile deployments.
His advice was to use proven infrastructure, event-driven architectures, queues, and observability tools, rather than reinventing the wheel. This, he said, allows teams to scale quickly while staying resilient.
Lessons from the Frontline
That philosophy shaped LinkedIn’s design. The LinkedIn Hiring Assistant was not built as a chatbot but as a structured hierarchy of agents. Each carried a specialised role—sourcing, calibration, email outreach—while a supervisor agent coordinated the flow.
“The way we architect our agent is almost like an org chart,” Ramgopal said. He explained that dividing responsibilities avoided overlap and ensured clear accountability.
Ramgopal stressed that the system’s strength came not from tools but from reusable skills. “Skills encapsulate the what and the how,” he said. These skills could be APIs, database queries, prompts or even other agents.
A skill registry allowed the assistant to discover and deploy the right capability in real time. “For the system to have agency, it must seamlessly discover the skills,” Ramgopal said. He added that this modularity made the system easier to evolve as new models or tools emerged.
Memory that Matters
Memory, he noted, was equally critical. To avoid the pitfalls seen in earlier AI systems, LinkedIn introduced a layered memory framework—working memory for sessions, long-term memory for recall and collective memory to capture organisational norms.
“From our early generative AI products, we learned you need world-class memory,” Ramgopal said. The result was sharper personalisation, fewer repeated errors and lower compute costs. Recruiters, in turn, received recommendations that felt tailored rather than generic.
Even with these advances, automation was never the end goal. The LinkedIn Hiring Assistant was designed with oversight at every stage. Recruiters could intervene at any stage of planning, calibration or final decision-making.
“Agents aren’t fully auto-fit as much as we’d like them to be,” Ramgopal said. “You have to design for humans in the loop.”
This philosophy underscored the larger theme of his talk. AI can handle the repetitive, tactical load, but people remain essential for judgment, nuance, and relationship-building. For recruitment, a field grounded in trust, that balance may well determine how widely such systems are adopted.
The post Can AI Make Recruitment More Human? LinkedIn Thinks So appeared first on Analytics India Magazine.
