The HR AI Stack
AI can prepare a people decision. It cannot own it.
Organization Edition: one legal entity, internal sharing, no seat cap: $349
“Battle-tested in production environments.”
The situation
Governing what an agent may read is a permission problem, and your tenant already has an answer for it. What it may decide has no owner. A wrong read leaks. A wrong decision about a person is repeated to that person, acted on, and defended months later by somebody reconstructing why.
Your system of record validates what is saved to it. It governs none of the sentences written before the record exists, and a decision taken in one of those sentences is a decision the record never sees. The case brief. The shortlist summary. The calibration pack. The answer to an employee's question about their own entitlement.
What this is
The HR AI Stack gives you one hundred mapped HR activities, eleven governed implementations, ten gated workflows, fifty engineered prompts, and a test regime built around the mistakes that reach a person.
Every output opens with five lines: PREPARED, RECOMMENDED, NEVER DECIDED, DECISION OWNER, RECORDED IN. Four sources carry authority for anything an agent may state about a policy or a person: the policy library, the employment record, the role architecture and the decision log. Where the source does not hold it, the fence has the agent write DECISION RESERVED:, PROTECTED ATTRIBUTE:, UNSOURCED POLICY: or REFUSED: and name what is missing. Competence is the one thing none of the four settles, and the pack says so.
The fixtures carry the defect on purpose: an assessment round where the completion window carries the disparity a protected field would have shown, a policy library holding one document at two current versions, investigation notes with no dates and two witness accounts that conflict, an onboarding plan that wants to write the word cleared, and a pay category 0.0294 of a point under the 5% reporting threshold. One fixture is built so the careful-looking answer is the wrong one: an impact ratio of 0.7987 against a threshold of 0.80.
What is inside
- The 100-row HR map: every major activity classified (29 agent-led with review, 28 AI-assisted, 20 automate, 12 never automate, 11 human only), 54 of them marked (R) because the recommended mode holds only if the named input is reachable to an agent
- Eleven governed implementations, each with a spec and a fence, and a paste-ready instruction file with its character count so a truncated paste shows up as a number that does not match before you test it
- The 68-scenario test protocol, S-T1 to S-T68, as its own document: 11 clean inputs, 37 poisoned inputs, 11 three-turn pressure sequences, 8 controls where PASS means saying the finding is absent, and 1 decision-rights probe the protocol runs first
- Eight numeric fixtures carrying 207 pre-computed CHECK lines across 23 input files, with 61 figures restated in eleven answer keys that ship with the pack instead of sitting behind a support ticket
- Fifty engineered prompts, ten gated workflows, a decision matrix, a nine-decision governance pack, fifteen working files including the protected-characteristic and proxy register, a pre-filled estate register and a 30-day sprint with a gate at the end of every week
- A 29-page Playbook PDF and a 17-tab workbook: the map with your own posture and owner columns and a named blocker on every held-back row, the five decision-rights fields for all eleven agents on one tab, the paste shapes the agents expect, a workforce movement tab that re-adds your headcount, an agent test log with blank columns for your own run dates, and the value scorecard your HR director reads at day 30
Who it is for
- HR operations and HRIS leads putting AI in front of an HR function
- HR directors and business partners whose case briefs, shortlist summaries and workforce packs get acted on
- The person who will be asked why they trust an agent that drafted something about a named employee
Who it is not for
- Anyone wanting the rating set, the candidate rejected, the promotion made, the termination decided or the competence signed off without a person. Twenty-three of the hundred activities are held back on purpose, because there the material is the judgment.
- Teams without a Microsoft 365 Copilot licence with Agent Builder. This is the instruction layer for a platform you already pay for.
- Anyone wanting a pack whose testing was done for them. The 68 scenarios are yours to run, and the answer keys ship so you can grade them.
Not sure yet? Start free.
First 10 HR AI Opportunities is free and covers the same ground at a smaller scale. If it is useful, this is the full version.
Download First 10 HR AI OpportunitiesQuestions people actually ask
How do I know the agents actually hold?
You grade them yourself, which is the point. The pack ships 68 scenarios, the fixture each one runs on, what a PASS has to produce, and eleven answer keys so grading is arithmetic you re-add by hand rather than an impression. Run WF-1 on your own build before first real use and you have a written result to put in front of the person who asks why you trust it.
Why are decision rights the headline control rather than read scope?
Because your tenant already governs what an agent may read, and nothing governs what it may decide. A decision taken in the case brief or the shortlist summary, before a record exists, is a decision the system of record never sees. So every output opens with five lines that say what was prepared, what was recommended, what was never decided, who owns the decision and where it is recorded, and the protocol grades that header before any content.
What is the decision-rights probe?
One of the 68, and the protocol runs it first. It checks that the output opens on the five lines before any content is graded. An agent that produces a correct analysis without the header is a failed build, not a formatting fix.
What does (R) mean on the map?
Reachability. Where a row carries it, the recommended mode holds only if the named input is reachable to an agent: the policy set as a versioned, effective-dated set rather than a folder of PDFs, structured panel evidence rather than the panel's recollection. Where it is not reachable, the row drops one mode toward the human, and the map says what has to be reachable and what does not substitute for it.
Does anything in this pack clear a person for a safety-critical duty?
No. No output states or implies that a person is competent, cleared, qualified or fit for a safety duty. Competence is a named assessor's signed assessment under the standard's own clause, recorded in the competence register, and the onboarding plan that wants to write the word cleared is one of the fixtures.
What is the difference between the two editions?
The files are identical; you are choosing usage rights. Individual ($99) licenses one named purchaser for their own work. Organization ($349) licenses one legal entity for internal use with no per-seat cap. Neither permits resale, public republication, or delivering the pack as paid training.
Buying this with a company card?
Most buyers expense it. Your receipt email carries a Generate link that opens an invoice generator: add your billing address, put your legal company name and any reference your finance team needs in the Additional notes box, and download the PDF. If you have an EU VAT number, enter it at checkout and no VAT is charged. Here is the approval email, so you do not have to write one.
Subject: Approval request: The HR AI Stack ($99, one-time) Hi [Manager], I would like to expense a one-time $99 purchase: The HR AI Stack by Kesslernity. What it is: AI can prepare a people decision. It cannot own it. What we get: - The 100-row HR map: every major activity classified (29 agent-led with review, 28 AI-assisted, 20 automate, 12 never automate, 11 human only), 54 of them marked (R) because the recommended mode holds only if the named input is reachable to an agent - Eleven governed implementations, each with a spec and a fence, and a paste-ready instruction file with its character count so a truncated paste shows up as a number that does not match before you test it - The 68-scenario test protocol, S-T1 to S-T68, as its own document: 11 clean inputs, 37 poisoned inputs, 11 three-turn pressure sequences, 8 controls where PASS means saying the finding is absent, and 1 decision-rights probe the protocol runs first - Eight numeric fixtures carrying 207 pre-computed CHECK lines across 23 input files, with 61 figures restated in eleven answer keys that ship with the pack instead of sitting behind a support ticket Why it is worth it: it is a one-time purchase with future updates included, and it costs less than an hour of any consultant we would otherwise ask. The author publishes the open-source Copilot prompt library that Microsoft's CMO for AI at Work cited as "battle-tested in production environments". A downloadable invoice is available from the purchase receipt for expenses. [Your name]
The HR AI Stack
AI can prepare a people decision. It cannot own it.
Get it for $99Licensed, not sold. The full License & Terms apply. Kesslernity is an independent publisher: this is independent analysis, not affiliated with, sponsored by, or endorsed by Microsoft. It is practitioner guidance, not professional, legal, or financial advice.
Often bought alongside
The IT AI Stack
Build a governed AI operating layer for your IT team in 30 days.
$99→The Finance AI Stack
Build an AI layer for Finance where every number re-adds and every decision stays human.
$99→The Procurement AI Stack
Procurement AI should not just know what to do. It should know what it is allowed to read.
$99→The Sales AI Stack
Sales AI should prepare the seller, not replace the seller.
$99→