AI with Michal

Team workshops

Workshop on Sourcing with AI

Private cohort workshop for sourcing teams and TA pods who own discovery and top of funnel. These pages spell out format, sample exercises, tooling, logistics, optional post-workshop engineering, and how to book.

All team workshopsAbout me

Agenda & samples

Example agenda and sample exercises

Timings shift with cohort size and depth. Most of the time is spent rebuilding real artifacts (anonymized) and writing guardrails your pod can reuse next week.

Modules (typical half-day / full-day)

  1. 1

    Operating model: draft versus decide for sourcers

    ~20 to 30 min

    Outcome: Shared vocabulary for assisted steps versus steps that need a named reviewer before anything external.

  2. 2

    Market map in one page

    ~35 to 45 min

    Outcome: Employer clusters, communities, and constraints a pod can defend in a standup without a slide deck.

  3. 3

    Parallel manual and AI-assisted discovery plans

    ~35 to 50 min

    Outcome: Two plans merged into a weekly execution checklist with owners, gates, and stop conditions.

  4. 4

    List building and enrichment boundaries

    ~25 to 35 min

    Outcome: Where enrichment helps routing, where it should never autopilot, and how to keep minimization honest.

  5. 5

    First-touch outreach: variants and policy pass

    ~30 to 45 min

    Outcome: Three variants with proof, one ask, and explicit flags before send.

  6. 6

    Handoff pack recruiters can use

    ~25 to 35 min

    Outcome: A short structured handoff: coverage, pools tried, outreach angles, unknowns, and next experiments.

  7. 7

    Weekly sourcing retro and metrics that are not vanity

    ~20 to 30 min

    Outcome: A retro format that documents experiments and sets next week's plan without invented pipeline math.

  8. 8

    Pod playbook handoff

    ~15 to 25 min

    Outcome: What to standardize Monday, what to spot-check lightly, what to revisit in 30 days.

Sample moments (marketing-safe)

Prompt frames are teaching scaffolding. Your team adapts them to your policies and templates. Do not paste real candidate data into a live classroom unless your security team signs off.

Market map for a new geography

Context

Leadership expanded a senior engineer search to a city the team rarely works. The sourcer needs a one-page map: likely employers, communities, visa or relocation realities, and three angles to try first.

Prompt frame (template)

Role: [title], metro: [city], constraints: [visa, comp band, remote policy]. Produce: (1) 8 to 12 plausible employer clusters with why each matters, (2) 6 non-obvious communities or meetups to research, (3) 3 angles for first-touch outreach that cite verifiable facts only. Flag stereotypes or protected-class proxies explicitly if they appear.

Quality checks

  • Employer clusters are specific enough to search tomorrow.
  • Communities are named, not vague regions.
  • Outreach angles avoid protected-class proxies and unsourced superlatives.

Before

A bullet list of big-brand logos copied from a generic template.

After

A tight map with rationale, research hooks, and outreach angles a sourcer can defend in a weekly standup.

Manual search plus AI: one weekly sourcing plan

Context

Half the pod writes tight, testable queries; half default to single-keyword search. Build two parallel plans, then merge into one weekly checklist with owners and verification steps.

Prompt frame (template)

Given [role brief], produce (A) three testable search strings with one-sentence rationale each, (B) an AI-assisted discovery checklist with 5 steps and stop conditions, (C) a merged weekly plan table with owner, day, and quality gate.

Quality checks

  • Search strings are testable and not comically long.
  • AI steps include stop conditions when results get noisy.
  • Merged plan fits a real week, not a fantasy sprint board.

Before

One bloated catch-all query and a separate generic ChatGPT thread nobody references.

After

One shared weekly plan that names who runs classical search, who runs assisted research, and when to merge results.

Outreach variants with a policy pass

Context

Reply rates dropped after obvious AI templates reached the market. Draft three variants, then flag anything that would worry legal or brand.

Prompt frame (template)

Rewrite this outreach draft in three variants under 90 words. Constraints: no salary promises, no guaranteed language, include one verifiable company fact, one clear CTA. Add a short policy flag list for the sender.

Quality checks

  • No unverifiable superlatives.
  • CTA is a single ask.
  • Policy flags are explicit even when the draft looks fine at first glance.

Before

A bloated template with six links and vague amazing opportunity language.

After

Three tight variants with one proof point, one ask, and a visible flag list before send.

Talent pool research with receipts

Context

The team needs alternate pools beyond the usual LinkedIn title search. Produce a short research memo sourcers can attach to a handoff.

Prompt frame (template)

Given [role brief], list 10 non-obvious pools (communities, meetups, repos, niche job boards). For each: why it fits, how to search it this week, and what would falsify the pool. Avoid stereotypes.

Quality checks

  • Pools are specific enough to act on.
  • Each pool has a falsify condition so the team moves on when wrong.
  • Stereotype and compliance risks are called out, not hidden.

Before

A keyword brainstorm with no owners or next step.

After

A prioritized pool list with search steps and falsify rules the pod can run in parallel.

Weekly sourcing retro tied to handoff

Context

Leadership wants more pipeline, but the team cannot explain what was tried. Build a lightweight retro format that connects search work to recruiter-ready notes.

Prompt frame (template)

Given [last week bullets], produce: (1) what we tried and evidence of coverage, (2) top 3 blockers with suggested owner, (3) 3 experiments for next week with success signals that are not vanity metrics, (4) a 5-bullet handoff for recruiters.

Quality checks

  • No invented funnel math; unknowns are explicit.
  • Experiments fit the team's real hours.
  • Handoff is readable in under two minutes.

Before

Paragraph updates that hide whether the issue was search, messaging, or downstream speed.

After

A short retro with experiments and a recruiter handoff that shows what was searched and what to try next.