Chapter 07 · Phase 4 of 8

Pilot the Seed Problem

Don't try to transform your entire workflow on day one. Find one personal pain point. Build one agent. Let the first success compound into the next.

1 · Assess 2 · Charter AI COEE 3 · Buy-in 4 · Pilot 5 · Deploy 6 · Train 7 · Govern 8 · Scale

This is the most important chapter for getting momentum right. When I asked Kristina Tsys how she got Fortive's AI program moving, she didn't describe a roadmap. She described a problem.

I started with my own challenge — I'm hiring across 10 different companies. Each company, different product, different solutions, different onboarding process, different background check. Every time I have a hire and I need to finalize it… previously, I had to go through multiple documents and pull out information while I'm on the phone. So my first agent was the onboarding agent.

Kristina Tsys · Technical & Engineering Recruiter, Fortive Listen →

One agent. One personal frustration. One concrete time savings. Then the next agent came easier. And the next. The model Kristina described — and the one I've watched work over and over in enterprise rollouts — is the seed problem. Find the seed. Plant it. Watch it grow.

What makes a good seed problem

Not every workflow is a good first candidate. The best seed problems share five characteristics:

It's personal. The pilot owner experiences the pain daily. Their motivation is real, not theoretical.
It's bounded. The workflow is narrow enough that an MVP can ship in 2–4 weeks.
It's measurable. Time saved per use, frequency per week, accuracy improvement — you can quantify it.
It's low-risk. No candidate-facing decisions. No compliance exposure. If the agent gets it wrong, the cost is a few minutes of re-work, not a discrimination lawsuit.
It's repeatable. Other team members face the same friction, so once it works, it spreads.

The seed problem menu

From every TA team I've talked to, these are the highest-yielding seed problems — the ones that almost always work as a first pilot:

Intake notes agent

Captures and structures notes from hiring manager intake conversations and candidate screens. Lets the recruiter stay present in the conversation. Kristina's team built this early. It's a near-universal win.

Onboarding info agent

An indexed assistant that holds every onboarding document, policy, and benefits doc and answers questions while the recruiter is on the phone. Especially valuable for multi-entity orgs.

Daily inbox / calendar agent

Summarizes inbox activity at the start of the day. Surfaces what's urgent vs. what can wait. Outlines the day's calendar with context. Kristina described this as "getting your personal assistant that is available 24/7."

JD drafting agent

Takes raw role information, prior-version JDs, and the hiring manager's intake notes and produces a structured, skills-based job description. Recruiter edits, manager approves. From 90 minutes to 15.

Sourcing brief agent

Given a JD and a market, produces a sourcing brief: target companies, search strings, market context, salary benchmarks. Patrick Lindsley moved from 2–3 hours of quality sourcing per week to nearly 10 with a workflow like this.

The recruiters winning with AI are using it to move faster so their judgment gets applied to decisions that actually matter.

Patrick Lindsley · Recruiting Operations Director Listen →

The 4-week pilot protocol

Tight timeline. Tight scope. Tight measurement.

Week 1 — Define

Pick the seed problem and the pilot owner
Document the current workflow (Matt Neylon's rule — "you can't automate what you can't write down")
Baseline the time, frequency, and accuracy
Identify the tool — for most orgs, this is the AI you already have a license for (Copilot, ChatGPT Enterprise, Claude for Work)

Week 2 — Build

Construct the agent or workflow inside the chosen tool
Test against 5–10 real cases
Iterate prompts and instructions until the output is consistently usable

Week 3 — Use

Pilot owner uses the agent for every applicable case for a full week
Log time saved, errors caught, friction encountered
One mid-week check-in with the AI COEE to surface blockers

Week 4 — Share

Pilot owner presents to the broader team — live demo, real data, real lessons
AI COEE captures the agent design as a reusable template
Decision: graduate to broader deployment (Phase 5), iterate, or kill

Pro tip

Pick the pilot owner carefully

The first pilot should run with someone who is genuinely curious and has credibility with peers. Not the loudest skeptic. Not the most senior. The person whose endorsement will move others. Brian Fink's "curious, empathetic, tenacious, mischievous" recruiter profile is the archetype.

Common pilot failures

Do

Start with one workflow, one tool, one person
Measure baseline before, measure outcome after
Document the prompt or agent design as a template
Demo live to the team — don't email a deck

Don't

Pick a candidate-facing workflow as your first pilot
Try to pilot 5 things at once
Skip the baseline so you can't prove the improvement
Let the pilot drag past 5 weeks

The compounding effect

The reason the seed-problem approach works isn't just psychological — though confidence does compound. It's structural. Each agent built creates a template. Templates accelerate the next build. Patterns emerge that the AI COEE can document and share. By the time you have three working pilots in three different recruiting roles, you've also built the institutional knowledge to deploy a fourth in days, not weeks.

That's how you go from one agent to a portfolio. And that's where Phase 5 takes us — the structured deployment from AI-Assisted to AI-Augmented to AI-Powered workflows.