playbook

How talent teams are really using AI at work, and what they stopped using

What talent teams are building with AI, where it saves time, what they stopped using and how to create your first repeatable agent.

By Talent Crunch - Berlin · 2026-09-22

How talent teams are really using AI at work, and what they stopped using

Findings from the AI in the daily workflow World Café, Talent Crunch and Greenhouse Software, September 2026

Eleven post-its landed under “saves time” and two under “tried and dropped”. Almost every win was built by the attendee themselves in minutes, inside tools their team already had, rather than bought by their employer.

Talent Crunch X Greenhouse_10.09.2026 ©Arundhati Shenoy_80 (1).jpg

Agents built in-house: three examples and their payback

An agent is a set of instructions we give an AI tool once so it runs the same task the same way every time. Set-up costs minutes, and the payback returns every time the task comes round.

Agent built

What it replaces

When it pays back

Weekly hiring summary for stakeholders

Assembling the same report by hand every Friday

Weekly, per stakeholder group

Job description drafted from intake notes

Writing the JD from scratch after the call

Every new requisition

Gender-neutral JD check with brand tone

Manual proofreading and inconsistent ad tone

Every job ad


Job description rewriting: one agent, more female applicants

One attendee built an agent that checks a job description for gender-neutral language and applies employer brand tone in the same pass. They reported more female candidates applying to those roles in a company that was already heavily male dominated.

One practitioner, one company, no control group - and a test any of us can repeat on our own requisitions this quarter


The full board: every AI use case recorded, by column

Talent Crunch X Greenhouse_10.09.2026 ©Arundhati Shenoy_31.jpg

Column

What was written

Detail

Using day to day

Pin

AI sourcing across 850M+ profiles in natural language, with agents for sourcing, screening and outreach

Using day to day

Juicebox (PeopleGPT)

Recruiting search across 800M+ profiles, described in plain language instead of boolean strings

Using day to day

Greenhouse Report Connector for Google Sheets

Pulls report data into Sheets and refreshes on demand, removing the export-and-paste cycle

Using day to day

German conversation tutor prompt

A long, detailed prompt used outside work for language practice

Using day to day

Getting the company to move

How do we make the company understand the importance of AI and accelerate implementation

Saves time

notetaker.ai and other AI notetakers

Intake meetings and interviews, including capturing interview feedback while it is fresh

Saves time

Weekly hiring summary to stakeholders

Agent assembles the week's activity and writes the update

Saves time

Job description from the hiring manager meeting

Agent drafts the JD from the notetaker's record of the intake

Saves time

JD gender neutrality check with brand tone

Agent rewrites the ad and applies employer brand voice in one pass

Saves time

Claude skills

Built once for job ads, playbooks and other repeat work, then reused

Saves time

Cowork skill creator and Cowork projects

Same purpose, building repeatable tasks once

Saves time

Copilot image creation

Mocking up an inter-department project so every team saw the same thing

Saves time

Outreach, CV scanning and screening

Volume tasks at the top of the funnel

Saves time

Translation, paraphrasing, summarising and analysing documents

Minutes to hours per document

Tried and dropped

AI outreach sequencer

Breached internal policies under GDPR in Germany

Tried and dropped

LinkedIn Hiring Assistant

Weaker at sourcing than the other AI tools tested


What we tried and stopped, and the reasons given

An AI outreach sequencer breached one attendee's internal policies under GDPR in Germany. AI tools ship globally at speed while local data protection and employment rules are not always built in, so the compliance review belongs at the start of an evaluation.

Recruitment AI also sits in the high-risk category under Annex III of the EU AI Act, with obligations for standalone systems applying from 2 December 2027 after the delay approved in June 2026.

Which of the tools in our stack would we have to declare?

LinkedIn Hiring Assistant went for a simpler reason: the attendee found it weaker at sourcing than the other AI tools they had tested.


The adoption blocker: uneven access and the internal case

The strongest post-it was a question: “How do we make the company understand the importance of AI and accelerate implementation?” Some attendees have changed how they work entirely because their employer handed them the tools, while others are behind through no choice of their own.

Bring leadership one working example, ask for a single tool to be tested, and set up a small task force to pilot it. Keep the claim modest.

Section's AI Proficiency Report, published in January 2026 from 5,000 knowledge workers in the UK, US and Canada, found 68% of AI-using employees save four hours a week or less.

Four believable hours will pass a finance review that a transformation promise will not


3 steps to your first agent

  1. Name one task you repeat every week and time how long it takes you

  2. Ask Claude to build the skill or agent for it, describing the task in your own words

  3. Refine it as gaps appear, then share it with your team so the saving multiplies


Frequently asked questions

What is an AI agent in recruitment?

A set of instructions given to an AI tool once so it performs a defined recruiting task the same way every time, such as drafting a job description from intake notes or writing the weekly hiring update.

Which AI tools do recruiters use day to day?

At this session: Pin and Juicebox for sourcing, the Greenhouse Report Connector for reporting, notetaker.ai for intake and interviews, and custom agents or Claude skills for job ads, translation and playbooks.

Can AI write job descriptions?

Yes, and the stronger use at this board was editing rather than drafting: running an existing JD through an agent that checks gender-neutral language and applies brand tone.

Is using AI in recruitment GDPR compliant?

It depends on the tool and how it handles candidate data. One attendee dropped an outreach sequencer because it breached their internal policies under GDPR in Germany. Run the compliance review before the pilot.

Does the EU AI Act cover hiring?

Annex III lists recruitment uses as high risk. Obligations for standalone systems apply from 2 December 2027 after the delay approved in June 2026, and GDPR applies now.


Sources

#ai #recruitment #talent teams #agents #world cafe #greenhouse