LinkedIn says its Hiring Assistant agent saves recruiters an average of 1.5 hours per role

The problem

Recruiting work mixes judgment with a lot of repetitive pattern-matching: finding candidates, checking each one against a job, and following up. LinkedIn wanted an AI agent to take on that repetitive work so recruiters could spend their time on the human side of hiring. The first version worked in English. The next problem was serving recruiters in other languages.

What they built

Hiring Assistant is an AI agent for recruiters. It is a team of specialised sub-agents (small AI components, each with one job such as sourcing or evaluating candidates) coordinated by a supervisor agent that plans the work and hands out tasks. Each recruiter has their own instance. It draws on LinkedIn's network of 1.2+ billion profiles. In June 2026 LinkedIn announced that the agent also works in French and German.

How they did it

LinkedIn describes the agent in two phases. First it chats with the recruiter to agree what the job needs. Then it runs sourcing and evaluation in the background, refreshes candidate pipelines daily and keeps people in the loop for decisions they are accountable for. For candidate evaluation LinkedIn trained custom models, which it says can judge candidates in seconds.

Adding French and German was harder than translating screens. The agent writes its text on the fly, so gender agreement, formal or informal address, capitalisation and number formats all have to be right in every step the recruiter sees. LinkedIn built a shared set of language rules (a rubric) written with linguists. It rewrites each English prompt into a language-aware version using those rules. For sub-agents that run on smaller models, it trains small per-language add-on modules called LoRA adapters on a shared base model. An AI judge, checked against expert linguists, scores output at scale, and linguists spot-check samples.

Results

LinkedIn states that recruiters are already saving an average of 1.5 hours per role with Hiring Assistant, and describes it as among the quickest-growing agentic products on the market. It states that the French and German launch extends the same productivity gains to recruiters in those markets. For quality, it reports that the AI judge agrees with expert linguists at a rate in the same range as individual linguists agree with the final label, without giving a number.