Where AI helps, and where it's a lawsuit
HR runs on text. CVs, job descriptions, interview notes, policies, performance reviews, exit interviews. Text is exactly what these models are good at, so the opportunity is real — and it comes with the tightest constraints of any role in this track, because HR decisions are legally protected and life-affecting.
The useful division isn't "AI good / AI bad." It's who does the deciding.
Genuinely good uses
- Rewriting a job description to cut exclusionary language and trim the requirements list to what the job actually needs.
- Summarising six interviewers' notes into the themes they agree and disagree on.
- Drafting first passes at policy documents, comms, offer letters.
- Answering "how much parental leave do I get?" from the handbook — with the passage cited.
- Structuring your own thinking before a hard conversation.
Genuinely dangerous
- Ranking or scoring candidates.
- Rejecting anyone automatically.
- Inferring anything about a person that isn't a demonstrated qualification.
- Anything touching protected characteristics. Ever.
The line, in one sentence:
AI drafts and summarises. Humans decide about humans.
And notice why. Not because a model can't produce a candidate ranking — it will produce one instantly, and it will look thorough, and it will have reasons.
That's the problem.