AI for HR: from hiring to training — what actually gets automated

01 — The niche

Why is HR a perfect — and underrated — niche for AI?

HR work is a conveyor of repeatable processes: vacancies, screening, interviews, onboarding, training, assessment, internal communications. Every process has an SOP and criteria — exactly what parametrizes well.

And the HR requests in the base go beyond "write a job post": "automate HR processes," "build career-guidance tools," even "how do I set up an AI-staff agency." People in this profession quickly see the point: AI isn't a text generator — it's a way to scale work with people without losing quality.

Here's how our [interview/onboarding/assessment] runs today,
step by step: [describe freely].
Pull out: 1) which steps repeat identically every time,
2) which criteria you keep in your head but never wrote down,
3) what could become a template today.

02 — The session cycle

How does session and assessment analysis get automated?

The most structured pattern from the cases is a three-step cycle around any session (interview, review, training):

• Prep — AI assembles a brief on the candidate/employee: inputs, history, questions matched to the meeting's goal;
• Assessment — after the session, notes or a transcript turn into a structured evaluation against your criteria;
• Strengths and growth areas — the output is a concrete development plan, not a vague "good job, keep trying."

This used to take an hour per person — so it was done shallowly or not at all. Now the depth of the review no longer depends on an HR person's free evening.

Candidate/employee: [resume or history, anonymized].
Meeting goal: [role assessment / performance review / growth plan].
Build a brief: 3-5 questions matched to the goal, what to watch
for in the history, where red flags might be, what to confirm live.

03 — Custom GPTs

What assistants does HR build for itself?

Per the cases, HR builds custom GPTs for specific processes: resume screening against role criteria, candidate email drafts for every stage, an onboarding navigator for newcomers, career-guidance tools.

An important observation from one participant: "AI raises the quality of your toolkit if you build for more than marketing" — meaning working tools for internal processes, not showcase bots. The same principle as everywhere: one assistant = one process with its own SOP, not a "universal HR bot."

04 — The principle

"AI is a first-grader": why is this the best mental model?

AI is a first-grader that needs clear instructions and micro-steps.— an HR expert from the focus group

This formula from the case is the best mental model for working with AI, period. You don't tell a first-grader "do it well" — you give small steps, examples, and you check the result. That's exactly how to set tasks for a model: micro-steps, criteria, samples.

Then comes the beautiful reversal: the HR people in the base noticed the same skill improves training human employees. Break a process into micro-steps, give examples, set criteria — that's what good onboarding is. Prompting turned out to be a gym for managerial clarity.

05 — The limits

What does HR not hand to AI?

HR's boundaries are about people and data:

• Decisions about people — hiring, firing, promotion are made by a human; AI prepares the material, not the verdict;
• Personal data — resumes and assessments carry sensitive information: anonymization or enterprise environments are mandatory;
• Bias — a model can inherit skews from your own examples; screening criteria should be checked for discriminatory patterns.

Same scheme as with lawyers: AI is the draft and the structure; the human is the decision and the responsibility.

Here are my screening criteria for role X: [paste the list].
Check: 1) are there proxies for gender, age, nationality here
(e.g. "graduate of a specific university" or "no gaps in
employment")? 2) which criteria should become direct,
measurable skills instead?

06 — Where to start

Where should HR start this week?

1. Take one cycle: say, interview debriefs
2. Formalize the assessment criteria (what you look at, what "good" is)
3. Prompt: "here are my interview notes [anonymized], here are the
   criteria — assemble an assessment: strengths, growth areas,
   recommendation"
4. Test on 3–5 past interviews, refine the criteria
5. Rules: decisions about people are yours; data is anonymized