Overview

Lecture URL: https://skill.re/learn/recruiting/designing-guardrails-what-ai-can-do-what-requires-human-approval.php

TRANSCRIPT: Designing Guardrails: What AI Can Do, What Requires Human Approval

Course: AI for Recruiters - Professional Credential

Module: Level 2: Hands-On Foundations

Section: Chapter 10 -- Guardrails and Supervised Use

Theme: guardrails-and-supervised-use

Lecture: 10.1

Duration: 60 min

Format: Workshop + Hands-On

Audience: Recruiters beginning to use AI tools

Prerequisites: L1 Certification

What you will learn: You'll build decision frameworks that specify which recruiting decisions AI

can support independently versus which require human judgment and approval. By the end, you'll have

clear guardrails for responsible AI use in your organization.

INTRODUCTION

AI is useful as a tool. AI is dangerous as a decision-maker. The difference is guardrails.

Guardrails are the lines you draw that say: "AI can do this autonomously. AI can help with this,

but humans decide. AI should never do this."

Without guardrails, AI creep happens. Drafting becomes sending. Summarizing becomes deciding.

Support becomes replacement. Clear guardrails prevent that.

Today, we're designing your guardrails framework.

THE THREE ZONES OF AI RESPONSIBILITY

ZONE 1: AI CAN DO AUTONOMOUSLY

These are tasks where AI output is genuinely useful and human review is optional (though good

practice).

Examples:

• Screening resumes for clear keyword matches
  - Summarizing interview notes
  - Drafting outreach messages (for human review/send)
  - Extracting structured information from CV
  - Research and background compilation

Why these are safe:

Output is transparent. Humans can easily see and verify what AI did. The consequence of error is

moderate (wrong summary, draft needs revision). These are decision support, not decision-making.

ZONE 2: AI SUPPORTS, HUMANS DECIDE

These are tasks where AI provides input, but humans make the final call.

Examples:

• Red flag identification (AI flags, human decides if it's legitimate)
  - Cultural fit assessment (AI provides observations, human decides fit)
  - Candidate comparison (AI organizes info, human decides who's stronger)
  - Concern evaluation (AI identifies, human evaluates)

Why humans must decide:

These involve judgment calls, fairness considerations, or company-specific context that AI can't

fully understand. AI is a powerful lens, but human judgment is essential.

ZONE 3: AI SHOULD NOT DO

These are tasks where AI shouldn't participate, even supportively.

Examples:

• Final hiring decision
  - Offer extension/rejection without human approval
  - Reference calls or candidate interviews
  - Determinations of discrimination or compliance issues
  - Decisions about accommodations or protected characteristics

Why humans only:

These decisions are too consequential, too legally sensitive, or too dependent on human judgment

and accountability.

BUILDING YOUR GUARDRAILS FRAMEWORK

To create guardrails:

STEP 1: INVENTORY YOUR RECRUITING WORKFLOW

What are all the tasks you do with AI? List them.

STEP 2: CATEGORIZE BY ZONE

For each task: Can AI do this autonomously? Does this require human judgment? Should AI not do

this?

STEP 3: DEFINE HANDOFF POINTS

Where does AI's work stop and a human takes over? Make it explicit.

Example handoff point:

"AI summarizes interview notes. AI output goes to hiring manager. Hiring manager reviews summary

and decides if candidate moves forward."

STEP 4: SET APPROVAL REQUIREMENTS

For decisions in Zone 2, who approves? What's the approval standard?

Example:

"For cultural fit concerns raised by AI, any two team members who read the assessment must

separately agree the concern is legitimate before impacting the decision."

STEP 5: DOCUMENT AND COMMUNICATE

Write down your guardrails. Share with your team. Make them explicit and consistent.

SOURCING AND OUTREACH

Zone 1 (AI autonomous): Research compilation, draft message creation

Zone 2 (AI supports): Candidate quality filtering, outreach prioritization

Zone 3 (AI should not): Final outreach approval (should be human)

SCREENING

Zone 1: Resume summary, skill extraction

Zone 2: Red flag and concern identification

Zone 3: Screening decision (human decides)

INTERVIEWING

Zone 1: Schedule coordination, note-taking support

Zone 2: Interview assessment, skill evaluation

Zone 3: Interview outcome decisions (human decides)

OFFER AND CLOSE

Zone 1: Offer documentation, acceptance tracking

Zone 2: Compensation analysis, counter-offer recommendations

Zone 3: Final offer decision, contract approval

ANTI-PATTERNS

ANTI-PATTERN 1: NO GUARDRAILS

Description: Using AI without any framework for what it can/can't do.

Why it fails: AI creep happens. What started as support becomes replacement.

How to avoid: Set explicit guardrails upfront.

ANTI-PATTERN 2: TOO RESTRICTIVE

Description: Keeping AI in Zone 1 for everything, missing leverage opportunities.

Why it fails: You're not using AI for decision support where it could be valuable.

How to avoid: Be honest about where human judgment is actually needed. Use AI for everything else.

ANTI-PATTERN 3: UNCLEAR HANDOFF POINTS

Description: AI does work, but it's unclear where human approval happens.

Why it fails: Work gets pushed forward without proper review.

How to avoid: Explicitly define handoff point for every task.

PRACTICE PROMPTS

Exercise 1: Inventory Your Workflow

List every task you use AI for in your recruiting. For each, is it Zone 1, 2, or 3?

Exercise 2: Define Your Guardrails

For each Zone 2 task, define: (1) What does AI do? (2) Who approves? (3) What's the approval

standard?

Exercise 3: Map Handoff Points

For your most critical processes, diagram: AI does X -> Human reviews -> Decision made. Where's the

handoff? Is it clear?

Exercise 4: Test Your Guardrails

Describe a scenario to your team: "AI flagged this candidate as a cultural fit concern. What

happens next?" If answers differ, your guardrails aren't clear.

Exercise 5: Document Your Framework

Write up your guardrails explicitly. Share with team. Make it canonical.

KEY TAKEAWAYS

1. Create three zones: AI can do autonomously, AI supports and humans decide, AI should not do.
2. Zone 1: Low-consequence support tasks where human review is optional
3. Zone 2: Decision support where human judgment is essential
4. Zone 3: High-consequence or legally sensitive decisions that require human accountability
5. Define clear handoff points. Where does AI's work stop and human judgment begin?
6. Set approval requirements for Zone 2 tasks. Who approves? What's the standard?
7. Document your guardrails explicitly. Communicate to your team. Make them consistent.

GLOSSARY

Guardrails: The lines you draw specifying which decisions AI can support independently vs. which

require human approval.

Zone 1: Tasks AI can handle autonomously. Low-consequence, transparent outputs.

Zone 2: Tasks where AI provides input, but humans make final decision.

Zone 3: Tasks AI should not do. Too consequential, too legal, too dependent on human accountability.

Handoff Point: The moment where AI's work stops and human judgment begins.

SYNTHESIS AND APPLICATION

Clear guardrails let you use AI confidently without worrying about AI creep. When everyone knows

what AI can and can't do, AI becomes a trusted tool, not a risk.

This week, build your guardrails framework. Start with sourcing/outreach. Define Zone 1/2/3.

Document handoff points. Share with your team.

REFLECTION EXERCISE

1. What recruiting decisions do you currently make without human judgment that you should be

getting approval for?

1. If you had clearer guardrails, what would change about how your team uses AI?
2. What's your biggest concern about AI in recruiting? Is it a guardrails issue?
3. How would you explain to executive leadership why certain decisions need human approval?

CLOSING REMARKS

Guardrails transform AI from a potential liability into a trusted tool. When your team knows the

boundaries, you can scale AI use confidently. In the next sections, we're building the systems and

culture that support responsible AI at scale.

AI for Recruiters Certification Program

Level 2: Hands-On Foundations | Guardrails and Supervised Use | Lecture 10.1

A SkillsClinic initiative.

Duration: ~60 minutes | Word Count: ~2,480