Ten HR AI implementations on two pages, ordered so the first week names nobody.
Each entry carries the row it comes from in the stack's 100-activity map, the mode it runs in, the difficulty, a deploy estimate, the impact and the named role that owns the decision its output feeds. Eight run agent-led with review and two are AI-assisted, both of them in hiring, because an agent drafting a job description is writing the instrument the round will be run on. The card also carries the eleventh implementation, with the date that funds it.
Three of the ten in full, and all ten named.
Two of the ten sit on rows the map rates 5 out of 5 and eight on rows rated 4, and the card says so rather than printing five stars ten times. The order turns on what a run has to be right against, not on the rating. The three below are the build week one is made of, which names nobody; the row the whole stack is built around; and the one that is gated on the round's impact ratio, because only the aggregate sees the disparity.
HR Workforce Brief BuilderAgent-led + review
The recurring pack from the agreed definitions, exceptions surfaced, every figure traceable to its source.
Map row 95 · Medium · about 3 hours · impact 4/5 · decision owner: HR operations lead
Employee Case Brief BuilderAgent-led + review
The evidence structured against each limb of the policy test, with what is absent named and conflicting accounts shown as conflicts.
Map row 34 · Medium · about 3 hours · impact 5/5 · decision owner: Decision manager, never the investigator
Candidate Evidence SummarizerAgent-led + review
Panel evidence per candidate per criterion, gaps named as gaps, gated on the round's impact ratio.
Map row 17, gated on 19 · Medium · about half a day · impact 4/5 · decision owner: Interview panel, for selection
The other seven, by name
HR Leadership Brief Builder · HR Policy and Knowledge Navigator · Policy Change Impact Analyst · Job Description Builder · Interview Preparation Assistant · Onboarding Plan Builder · Learning and Development Needs Analyst
On the card each of those carries the same six fields as the three above: the map reference, the mode, the difficulty, the deploy estimate, the impact and the decision owner.
What the deploy estimates assume
Estimated build time for one experienced HR operations or HRIS analyst following the stack's implementation guide, excluding organizational approvals and excluding your own prep: the policy register, the metric definitions, the criteria as structured fields, and your protected-characteristic list. The prep is usually the longer half, and the card says so. These figures are estimates from the same build pattern on comparable agents, timed while building the earlier stacks in this line. Time your first build and replace them with your own number, which is the one your next ten builds should be planned against.
The highest-impact rows on the map are not the first ten, and the gap is the point.
Termination, promotion, performance ratings, individual pay, candidate rejection, applying redundancy criteria to named people, naming a successor, competence sign-off for a safety-critical duty and releasing a payroll are all rated 5 on the map, and an implementation built on any of them would contradict the map that authorises it. Twenty of the twenty-seven highest-impact rows sit in HUMAN ONLY or NEVER AUTOMATE.
On the full card that ships in the pack, five of the ten name the restricted row that sits one step past them, because knowing where the agent stops is what makes the output usable by someone who has to defend it. A restricted row is not a gap in the product.
Nothing here prepares a safety authorization either. No output states or implies that a person is competent, cleared, qualified or fit for a safety-critical or regulated duty, and an onboarding plan that writes the word cleared has signed a certificate nobody assessed.
highest-impact rows on the 100-activity map sit in HUMAN ONLY or NEVER AUTOMATE. In HR the highest-value work and the least delegable work are largely the same work, so the card holds two rows rated 5 and eight rated 4, and says so.
Ordered so the first week names nobody.
The sequence turns on one property: what a run has to be right against. An agent checked against a library with no version, no effective date and no owner cannot be graded at all, so each source lands a week before the agent that reads it, and the builds that sit next to a safety clearance, a budget or a public filing come last.
Week 1, the week that names nobodyTwo builds
The workforce brief and the leadership brief, plus the policy library. Both report on populations and neither names an individual, which makes week 1 the cheapest place to find out how your own build behaves. The week's long job belongs to the library owner: a version, an effective date and a named owner on every document week 2 will cite.
Week 2, the policy weekThree builds
The Navigator and the Policy Change Impact Analyst run on the library you versioned in week 1. The Case Brief Builder is the first agent in the sprint that names a person, and it is the row the whole stack is built around.
Week 3, the hiring weekThree builds
A description, the panel's questions, and the round. Every summary in a round can be accurate and the round still show disparity, and only the aggregate sees it, which is why the summarizer is gated on the impact ratio.
Week 4, the week where being wrong costs mostThree builds
The onboarding plan, the needs analysis and the pay gap preparer, plus the day-20 scorecard. These three sit next to a safety clearance, a budget and a public filing, and none of them makes any of the three. If the calendar is tight, cut implementations from week 4 rather than cutting a gate.
Your policy library is the bill.
Four of these ten depend on the approved policy set being versioned, effective-dated and readable by an agent. If yours is a folder somebody maintains personally, versioning it is the real cost of month one. The second bill is smaller and sharper: the published essential criteria as structured fields rather than advert prose, and your own protected-characteristic list, which is your counsel's call and not ours.
It is on the free card and it is on the paid page for the same reason: a buyer who discovers it in week four is right to be annoyed. Week 1 is sequenced to deliver two agents that need no library at all, so it delivers two working agents while the library work is still running.
of the first ten builds read the policy library, and none of them is in week one. The work is writing down which policy is current, at which version, effective from when, owned by whom.
Not one of the ten. Sequenced for a calendar that belongs to somebody else.
Directive (EU) 2023/970 on pay transparency had to be in national law by 7 June 2026. Under Article 9, employers with 250 workers or more file a first report by 7 June 2027, employers with 150 to 249 workers by the same date and then every three years, and employers with 100 to 149 workers by 7 June 2031. Every report covers the previous calendar year. Read those two facts together: for employers with 150 or more workers the first report is due 7 June 2027 and it reports calendar year 2026, so the figures in it are being created by payroll now.
Pay Gap Reporting PreparerAgent-led + review
The statutory figures to the prescribed method, with every category tested against the 5% threshold before the report is submitted rather than after.
Map row 65, gated on 64 · Medium, on top of a data join that is a project in itself · impact 4/5 · decision owner: Reward lead, with employment counsel
Condition (a) of Article 10(1) is arithmetic, and the agent prints each category difference to enough decimal places to resolve the distance to 5, never rounded onto the threshold. Condition (b) is a legal position taken by a named person: the agent never decides that a gap is justified and never proposes what an individual is paid. It is the first thing to cut if the data join is not ready, because cutting it costs you a filing you were going to prepare by hand anyway, and cutting a gate costs you the reason any of this is safe to circulate.
Free. Pay $0 at checkout and the PDF lands in your library.
Licensed, not sold. Free resources ride the same License & Terms as everything else here (Article 3.1); the full notice is at the foot of this page.
Practitioner guidance, not legal, employment-law or medical advice. AI prepares a people decision. It cannot own it.
Two pages, PDF. Pay what you want if you like it, starting at $0.
The card is the sequence. The Stack is the build.
The card is the two-page table, the week-by-week order, the month-one cost and the list of what the ten leave out. It is a sequence, not a build. The Stack is where all eleven arrive as paste-ready builds with their decision rights and their refusals written down, and where the fixtures and the answer keys live so you can grade your own runs by hand.