Back
  • 2026
  • AI product
  • Agent workflows
  • B2B

An AI recruiter humans can steer

Origin

At XING, Fabi began as a candidate-side AI assistant: a way to bring more context into job search. After leaving, I revisited the idea from the other side of the marketplace — asking what AI could do for recruiters, not instead of them, but with them in control.

My role

I shaped the 2024 assistant concept, candidate journey and AI response design, acting as the link between frontend, data engineering and data science as structured AI output became interface. In 2026, I independently extended the idea into a recruiter-facing suite.

Where it stands

At XING, we set the original Fabi assistant aside to keep job search focused. This case starts with that work and then moves into my independent recruiting concept — shaped by the question I kept coming back to: how can AI support recruiting without taking judgement away from people?

Two annotated Fabi mobile concepts with salary and regional demand charts, skill and location filters, company cards and suggested follow-up questions
Our first AI job-search assistant, 2024

Inspired by Amazon’s Rufus, we explored guidance at the point of search: turning a job query into context on roles, salaries, demand and skills.

Engineering chat excerpts and annotations comparing JSON and Markdown response formats, distinguishing missing from insufficient data, and inspecting available profile inputs
Product experience starts in the backend

I worked hand in hand with data engineering, data science and frontend to challenge rigid response formats, understand available data and define how Fabi should handle uncertainty. Because some of the most important design decisions happen long before anything reaches the screen.

Matrix of Fabi response states showing how salary and skills answers change when job or profile data is available, insufficient or missing
Designing for AI uncertainty

I mapped every combination of available, missing and insufficient data, then defined when Fabi should answer, qualify its response, ask for more context or hold back.

What if the most valuable AI in job search wasn’t for searching at all?

I shifted the idea towards recruiting: an agent prepared with candidate context, able to handle early qualification so recruiters could focus on deeper conversations.

Three candidate-facing Fabi phone screens showing job preferences, confidence levels and role comparisons, alongside connected recruiting workflow nodes
Fabi, reimagined for 2026: From search answers to first conversations.

I revisited the idea independently, this time from the recruiter’s side: a voice agent that could handle early qualification and make more room for deeper human conversations.

Recruiter-feedback slide grouping four areas where AI could support hiring while keeping human judgement.
The biggest recruiter problem was time

Recruiter feedback showed that AI’s strongest value was not replacing judgement, but reducing the manual effort around finding, matching and preparing candidates.

B2B AI research slide summarising recruiter feedback on where AI could support hiring without replacing human judgement
The solution: scale preparation, preserve judgement

Fabi connects search behaviour, profile data and consent into a qualifying pre-call — entering the conversation already prepared with the candidate’s profile, answers and intent, so recruiters receive fewer cold leads and clearer next steps.

Fabi’s desktop workflow editor with connected nodes for greetings, candidate questions, qualification, recruiter handover and notifications
Fabi’s Add a node dialog with options for conversation, information collection, recruiter handover, call routing, notifications and ATS updates
Complete Fabi call workflow showing branches for strong, potential and weak matches, recruiter handover, notifications and call endings
Phone displaying Fabi’s workflow editor with connected qualification nodes, an Ask AI field and a Publish flow button
Orchestrate the agent. See every decision.

I designed the workflow editor to put recruiters’ expertise into qualification and handoffs. Coloured nodes, visible routes, plain-language conditions and few but meaningful icons make that logic easier to inspect.

Fabi’s recruiter dashboard with activity metrics, conversation trends, common candidate questions, high-fit prospects and conversations needing human review
See the whole operation. Human attention where it matters.

I brought conversation trends, candidate questions, promising prospects and cases needing review into focus. Match signals show why someone stands out; restrained colour helps recruiters see where to act.