Hiring Assistant - shaped by customers, powered by AI innovation

Editor's note: This article originally appeared on LinkedIn

Since we announced Hiring Assistant last year, results from our limited launch with select customers have been truly inspiring. Recruiters are saving over four hours per role, reviewing 62% fewer profiles before reaching a confident short list, and seeing a 69% improvement in InMail acceptance. 

Today we’re excited to share that Hiring Assistant, the only AI agent for recruiters powered by the world’s most dynamic talent network, will be globally available in English by the end of September. Since our initial announcement, we’ve spent a lot of time listening to customers. They wanted a more natural, responsive, conversational way to engage with the assistant, deeper integration with their existing tools, and greater transparency into how the Hiring Assistant thinks and reasons.

Here is a closer look at how direct insights from our customers have shaped the product evolution.

What’s new: a more agentic, conversational and connected experience

When we first introduced Hiring Assistant last year, we focused on providing assistance through the core steps of the sourcing workflow. Today, we’re taking a major step forward - we've expanded the product to support more natural, conversational interaction across more parts of the hiring process.

• Making Hiring Assistant more agentic, responsive and conversational: We’ve made it easier for recruiters to describe needs in their own words and work alongside Hiring Assistant to find the right candidates. We did this by re-engineering our agent orchestration and tool-calling systems so conversations feel fast and fluid, while still allowing for deeper reasoning which leads to better outcomes. Hiring Assistant now shows visible progress as it works, allowing recruiters to step in and guide it as necessary, balancing speed with the thoughtful analysis they expect.
• Automating pre-screening: Hiring Assistant will engage interested candidates in pre-screening conversations via InMail to collect basic pre-screening details (like location preference and willingness to relocate) or verify key requirements set by the recruiter. As part of pre-screening, Hiring Assistant can also answer basic candidate questions on the role or the company. It then summarizes takeaways with transparent reasoning and grounding, helping recruiters focus on the candidate conversations that matter most.
• Accelerating applicant evaluation: Hiring Assistant uses the LinkedIn profile, application resume, and answers to screening questions to evaluate LinkedIn applicants against recruiter-defined qualifications. This lets recruiters quickly find top applicants and get fact-based summaries of how qualifications match the role.
• Connecting Hiring Assistant to your ATS (Connected Projects): And soon, with our ATS (applicant tracking system) integrations, recruiters will be able to evaluate all of their applicants with Hiring Assistant, whether they applied on or off LinkedIn. They’ll also unlock bidirectional sync of candidate evaluations, candidate stage updates, and more, between LinkedIn and their ATS, so their teams have full context. We strongly believe that enterprise agentic products should offer workflow cohesiveness, and this advancement reflects our commitment to delivering seamless connectivity across recruiting systems.
  

The tech powering Hiring Assistant: delivering speed, personalization, and quality

We’re committed to building tech that fits naturally with how teams operate. To help recruiters reliably find the right candidates, we’ve made under-the-hood updates to the Hiring Assistant’s agentic workflow, quality assessment, memory, and much more. 

All of this is built on top of the LangGraph framework to be fully agentic through multi-step orchestration. This enables deeper reasoning and execution so the assistant can fully understand and fulfill the recruiter's intent. Hiring Assistant doesn’t just listen – the product asks clarification questions, offers recommendations, and adapts based on context. We modularized key cognitive components and refined each for stronger contextual memory, more nuanced understanding, and more precise execution. Key advancements include:

• Principled multifaceted quality assessment: We built a comprehensive quality evaluation system that leverages a combination of expert human judgements and the latest reasoning LLMs to evaluate the Hiring Assistant’s performance. This system allows us to automatically measure (1) how well the agent is following product policies, and (2) how well its performance matches human expectations, even before releasing new updates into the product. We also improved the agent based on feedback from our customers, which we gather both explicitly through thumbs-up/down feedback on content generated by Hiring Assistant, and implicitly via in-product actions such as contacting or archiving candidates. We expect these quality improvements to accelerate as we make Hiring Assistant available to more customers.
• Improved cognitive memory: Personalization is at the core of Hiring Assistant’s evolution. Over the past few months, we rebuilt our memory architecture; improving how data is organized, processed offline, and retrieved online. For example, if a recruiter consistently prefers candidates from fast-growing startups with specific skill combinations, the Hiring Assistant remembers this pattern. Instead of repeatedly asking, it automatically surfaces candidates who fit that preference, while still allowing the recruiter to adjust or override. This memory-based cognition creates more intuitive and personalized interactions, turning the Hiring Assistant into a partner that learns preferences and evolves with each recruiter over time. 
• Proprietary LLM fine-tuned for recruiting: Off-the-shelf LLMs are powerful, though not optimized for recruiting. Over the past few months, we’ve built and refined our own fine-tuned model for this domain. Using human-annotated evaluation sets, model ensembles, LinkedIn’s rich Economic Graph data, and advanced training techniques, we’ve consistently outperformed general-purpose models. Our quality framework also ensures we can seamlessly integrate improvements from new commercial LLMs while maintaining our differentiation and edge.
• Proprietary model-driven semantic retrieval: We added LLM-based semantic retrieval as part of sourcing capabilities. This allows recruiters to source candidates with queries they could not previously express using facets and keywords. This, in conjunction with our agentic candidate evaluation, can help recruiters discover a broader range of relevant candidates - a capability resonating strongly with our early customers. 
• Responsible AI by design: As part of our unwavering commitment to responsible AI, we rigorously test for risks such as hallucinations and bias, and design our workflows so that recruiters always remain in control. Every action the assistant takes is transparent and open to feedback, helping the system continually improve with real-world use.
  

What’s next for Hiring Assistant 

We’ve loved seeing early reactions to Hiring Assistant and watching how our updates to conversational workflows, graph‑based planning and execution, and deeper personalization translate to recruiter success.

As we celebrate this milestone, we remain committed to empowering recruiters. To learn more, visit this page and check out Mark Lobosco’s video on how early adopters are seeing success with Hiring Assistant.