Samy Aumar built a working AI agent for Qonto's recruiting team in an afternoon, without engineers, that takes a recruiter from intake call to live job posting in under a day. The agent is in production. The team uses it daily. The point of the conversation was to prove this is not exotic; it is a practitioner move available to anyone willing to spend a few hours architecting.

Samy Aumar (People Systems Excellence Manager at Qonto) joined Nolan Church and Siadhal Magos on 10x Recruiting (more episodes on the 10x Recruiting hub) for the kickoff of the "How I AI in recruiting" series. It features practitioner show-and-tells from teams who have moved AI from experiment to workflow infrastructure. The episode covers Qonto’s AI mandate, the intake-to-job-posting agent that compresses a multi-day workflow into hours, and the market intelligence agent that turns competitive research into a 30-minute task. It also gets into the no-engineers-needed reality of modern agent builders, and why B-minus outputs beat blank pages every time.

This recap is the cheat sheet: what to build first, how to set quality floors across your team, and the operating model behind it all. That model separates the recruiting teams using AI to compound output from the ones still treating it as a side experiment.

AI mandates beat AI experiments

Qonto issued an internal AI mandate six months into the partnership with Metaview. The mandate was simple: the expectation across the recruiting org is that every individual is figuring out how to use AI for use in their daily workflow. That expectation, set top-down, is what separated Qonto from teams where AI is a side project owned by one curious recruiter.

The teams that treat AI as optional get the variance you would expect. Sophisticated users get massive use; everyone else stays on the margins. The mandate flattens the variance by making "I haven't tried it yet" structurally untenable. It does not mean the company prescribes a tool; it means the company prescribes the expectation of use.

The cultural signal matters as much as the operational one. The recruiter who joins Qonto knows on day one that AI fluency is part of the job description, not an extracurricular.

No engineers needed

The single biggest blocker for most recruiting teams is the story that AI agent setup requires an engineering investment. Samy's evidence destroys that story. The intake-to-job-posting agent he demoed live was built in an afternoon. The second one took 15 minutes.

The current tooling (Metaview, Dust, ChatGPT custom GPTs, and similar agent builders) is genuinely no-code. Recruiting ops people who are willing to spend a few hours can architect production workflows. The bottleneck has moved from technical capability to thinking--about-the-workflow.

That shift is good news for recruiting leaders. It means the people closest to the workflow are the right ones to build the agents. The DevOps queue is no longer the gate.

B-minus outputs beat blank pages

The reframe that lands hardest with Nolan is this one. He used to start every job description from a blank page; the cycle time was days. Now the cycle starts at an 85% draft from a custom GPT and lands at a finished job description in under an hour.

The perfectionism trap kills more AI adoption than any other failure mode. If the team is waiting for the agent's output to be flawless before using it, they will never use it. The right benchmark is whether the AI's output is better than starting from zero. For 95% of recruiting tasks (job specs, market research, candidate outreach drafts, debrief summaries), the answer is yes the first time.

The 15 minutes of human review and polish is where the recruiter's judgment shows up. The agent does the structural work; the human adds the company-specific intelligence the agent does not have.

Qonto's intake-to-job-posting agent

The flagship example Samy demoed: a workflow that takes an intake-call recording, runs the audio through Metaview Notetaker, and pulls the structured output into a Dust agent. From there it generates a draft job specification, runs it through Qonto’s brand voice and DEI guardrails, and outputs a final job description ready for ATS posting.

End-to-end cycle: under a day. The recruiter spends the morning on intake, the afternoon on review, and the role is live by close of business. Compare to the legacy cycle: an intake call, then waiting four days for the hiring manager to send back a JD doc, then a rewrite, then internal review, then posting. That cycle routinely takes a week to ten days.

The compounding benefit is the rest of the time the recruiter now has back. They can run two more intake calls in the same week, or invest the saved hours in candidate relationship building. Either way, the throughput of the function goes up without headcount.

The market intelligence agent

Samy's second demo: a Market Intelligence agent that consumes a role description and outputs a comparative analysis across competitors' postings for similar roles. Comp benchmarks, must-have-skills patterns, language conventions, and DEI commitments are all surfaced in minutes. The recruiter walks into the hiring-manager kickoff with the competitive context already in hand.

This is the work that used to be either skipped entirely (most teams) or assigned to a recruiter ops analyst (the well-resourced teams). The agent compresses what was a 4-hour research task into a 30-minute review. The hiring manager experience flips: the recruiter is no longer asking "what do you want" but presenting "here is what the market is doing, where would you like to differentiate."

That posture change is most of why the function gets taken more seriously inside Qonto. The recruiter shows up as the market expert, not the requirements collector.

Recruiting ops as quality architect

Siadhal's reframe lands here: the role of someone like Samy is not just to build agents for himself. It is to architect quality across the whole team. A well-built agent published to the team becomes the new floor for what good looks like at Qonto.

The implication is structural. Letting every recruiter discover their own AI workflow produces a "thousand flowers bloom" variance problem. Centralizing the best version of each workflow and making it available org-wide closes the gap between the sophisticated users and everyone else. The mandate plus the centralized agent library is how the variance gets killed.

This is also where recruiting ops earns its seat. Architecting AI quality is the highest-use thing a recruiting ops function can do in 2026. Pipeline reports and ATS hygiene are infrastructure; quality architecture is strategy.

Where AI gives recruiting teams use

Samy's stack is illustrative of the broader pattern across teams getting real use. The common ingredients: capture every interview signal, build agents on top of the captured signal, set quality floors, and run the workflows centrally rather than ad hoc.

Metaview Notetaker captures the intake call and the candidate interviews so the conversation becomes structured input for downstream agents. AI Sourcing generates targeted candidate lists per role criteria. Application Review handles inbound volume so the senior team can focus on the work where their judgment matters. Reports surfaces the patterns across interview data that close the loop on quality. For the AI-augmented recruiter pattern in depth, see claude-for-recruiters, and for the interview quality layer that feeds the whole stack, see good-interviewer-bad-interviewer.

Numbers from Metaview's 2026 AI & Hiring Alignment Report, surveying 505 recruiting leaders and hiring managers across North America and EMEA. The 55% vs 35% vs 21% gradient is exactly the Qonto thesis in chart form. Depth of AI adoption is what predicts the outcome, not whether AI is present at all. Teams that mandate AI as core to the workflow operate in a different quality tier than teams that experiment on the margins.

The operating shift

Three concrete moves any recruiting leader can run on Monday:

One: issue the mandate. Tell the team the expectation is everyone is finding ways to use AI for use in their workflow. Make the expectation visible, durable, and tied to growth-plan conversations. Without the mandate, you get variance; with it, you get compounding adoption.

Two: appoint the architect. Pick one person on recruiting ops to own the central agent library. Their job is to build the best version of each workflow, publish it to the team, and continuously upgrade as the tooling improves. The architect is how quality scales without you personally vetting every output.

Three: ship the first agent in a week. Pick the most painful workflow on your team (probably JD writing or candidate-debrief summarization) and ship the first agent in seven days. The agent does not need to be perfect; it needs to be live. Iteration starts after the team is using it, not before.

The teams that internalize these three moves compound output the way Qonto has. That is the operating shift.