HR — Charter

HR owns the people system: who joins, how they develop, how performance is judged, and how the organisation behaves. In an AI-operated company it also owns something new and uncomfortable — the honest conversation about what automation changes for the people already here.

Handled well, that conversation is why automation succeeds. Handled badly, it is why people quietly route around it.

What this role owns#

Hiring. Attracting, assessing and onboarding, fairly and lawfully.

Capability. What skills the organisation has, what it needs, and how the gap closes.

Performance. How contribution is judged, and how that judgement is made consistent.

Employment obligations. Contracts, working conditions, and the rules that apply when automated systems touch employment decisions.

Change management. The people half of every automation programme.

KPIs#

Where AI touches employment — carefully#

This is the most legally sensitive automation in the company, and the rules are tightening.

Under the EU AI Act, AI used in employment, worker management and access to self-employment is classified high-risk. Those obligations were deferred to 2 December 2027 by the Digital Omnibus package approved on 29 June 2026 — deferred, not removed. Meanwhile transparency obligations applied from 2 August 2026: candidates must be told when they are interacting with an AI system.

The practical position:

• Screening that rejects is the high-risk case. Ranking or summarising for a human who decides is a materially different posture.
• Test for disparate impact before deploying, not after. A model trained on past hiring learns past hiring, including its biases. Removing a protected attribute does not fix this, because other fields predict it.
• Be able to explain a decision. "The system scored you low" is not an acceptable answer to a candidate, and increasingly not to a regulator.
• Keep a human decision-maker on anything adverse.

Position as at 2026-08-04, and not legal advice. Obligations depend on your jurisdiction.

AI agents in this function#

Job description assistant — drafts from a role brief, checks for language that narrows the candidate pool unnecessarily.

Application summarisation agent — summarises applications against stated criteria for a human reviewer. Summarises; does not rank or reject.

Scheduling agent — interview logistics. Mechanical, high-volume, zero judgement.

Onboarding agent — provisions access, tracks completion, answers routine questions.

Policy question agent — answers "how much leave do I have" and similar from the handbook.

What stays human: every hiring decision, every performance judgement, every disciplinary matter, every termination, and any adverse outcome. Not because a model could not produce a recommendation, but because someone must be accountable to the person affected.

SOPs#

• Hiring — structured, same questions, scored against defined criteria. Structure is what makes hiring fair and defensible, with or without AI.
• Onboarding — access, equipment, training, and a named buddy. Measured on whether the person is productive, not on whether the checklist completed.
• Performance review — cadence, evidence, and calibration across managers.
• Off-boarding — access revoked same-day, knowledge captured before the last day.
• AI-in-employment review — any system touching an employment decision is assessed before use and re-tested periodically.

Templates#

Job description · structured interview scorecard · onboarding checklist · performance review record · SOP format for people processes.

Workflows#

In: hiring requests · applications · performance cycles · exits · capability gaps from department leads.

Out: hires · capability plan · performance outcomes · retention analysis · policy positions.

Handoffs: Legal and Compliance for employment obligations · Finance for headcount cost · COO for capacity · every department for hiring and capability.

FAQ#

Will AI reduce our headcount?#

It changes what people spend time on more than how many you need. Be honest about it either way — people can tell when an automation programme is described as "freeing you up" while roles quietly disappear, and that is how trust is lost for every future change.

Can we use AI to screen candidates?#

To summarise and organise for a human, yes. To reject automatically, treat as high-risk: test for disparate impact first, keep a human decision-maker, disclose it, and be able to explain any adverse outcome. In several jurisdictions this is a legal requirement, not a preference.

How do we get people to adopt AI tools?#

Solve a task they already find tedious and let them see the time back. Low adoption is almost always a signal that the tool does not fit the actual work — treat it as feedback, not resistance.

What is the biggest people risk in an AI-operated company?#

Skills quietly not forming. If nobody ever does the work from first principles any more, the organisation loses the ability to judge whether the automated version is right — and that becomes visible at exactly the wrong moment.

What else is coming for HR