For CHROs and People leaders
Start with one document—not a company-wide rollout.
Use a document HR already owns to make AI contribution, human review, discussion, and approval visible from the first draft to the official version.
PDF · 7 pages · roles, workflow, review gates, and a one-document checklist
One document. One working version. Named reviewers and a named decision owner. No prompt monitoring.
01 · Where the productivity goes
A habit this size does not stay at home. It walks into work with the people who have it—and access alone has not moved the result. Redesigned work has.
ChatGPT reaches more than one billion people every week, and the familiar chat is turning into a system that researches, uses connected tools, takes multi-step action, and repeats work on a schedule. The gains show up at the desk and then stop: McKinsey's 2026 survey put the share of organizations reporting AI contributing to EBIT at 37%, against 80% reporting individual productivity gains, essentially unchanged year over year. The exception is instructive. AI high performers—about 6% of respondents, reporting at least 5% EBIT impact and significant value—worked differently: nearly three-quarters had fundamentally redesigned workflows, against about a quarter of everyone else.
02 · The workplace
Policy did not stop adoption. It pushed part of it outside approved systems.
Access controls cannot manage AI use that sits outside approved systems. Redesign the work, name a human owner, and support people without monitoring their prompts.
66%
03 · The working system
Employees now work with personal AI assistants, AI agents assigned to team workflows, and company-governed AI agents across functions. The formal chart records people and reporting lines—not each contribution, review point, handoff, or decision owner.
What the org chart records
People, roles, and reporting lines
CHRO / Head of People
How work gets done
The formal team—and the larger working system
One person
Supports one employee's own work.
Accountable: employee
One team workflow
Performs defined steps in a shared process.
Accountable: workflow owner
Across teams
Works across approved functions or systems.
Accountable: executive sponsor
Map the work—not employees' private prompts.
A named person remains accountable for every consequential decision.
AI contribution is not the problem. Unclear review and decision ownership are. Start with one workflow: who contributes, who reviews, and who decides.
57%
of employees reported nontransparent AI use at work
University of Melbourne + KPMG · employees in 47 countries · 2025That is a reason to clarify roles and review—not to monitor private prompts.
04 · The human factor
This is not a discipline problem. It is a design problem: what gets checked, by whom, and before what moves forward.
2 in 3
In the same global study, reliance often ran ahead of verification. A review rule cannot stay implicit when fluent output looks finished.
University of Melbourne + KPMG · employees in 47 countries · 202558%
It held across ten work tasks: planning, sequencing, deciding. People who modified, challenged, or rejected what the AI agent produced reported more confidence in their own reasoning. People also described ideas that no longer felt like theirs.
Technology, Mind, and Behavior (APA) · 1,923 weekly AI users, US + Canada · 10 simulated work tasks · 20261h56m
Researchers at BetterUp Labs and Stanford's Social Media Lab named it workslop. 41% of US workers surveyed had received it; about half thought less of the colleague who sent it, and 42% trusted them less. The estimate at scale is roughly $186 per affected employee per month.
HBR · BetterUp Labs + Stanford Social Media Lab · 2025For most people ChatGPT stays a tool. For a small group it does not. OpenAI and MIT found emotional engagement concentrated among heavy daily users. Across the study, longer daily use was also linked to more emotional dependence and problematic use.
OpenAI + MIT Media Lab · 2025No covert diagnosis. HR should not infer mental health from prompts or label employees. It should create clear stop conditions, human review, manager guidance, and a safe route to support when observable work or wellbeing concerns arise.
05 · The ownership gap
AI access is not transformation. Licenses, integrations, and policy are necessary; they do not redesign the work.
52%
of organizations do not involve HR directly or cross-functionally in AI strategy and vision
SHRM State of AI in HR · 2026Is HR a named co-owner of your AI transformation?
How 8Hats Lab helps
We turn the evidence into operating practice: a clear HR mandate, redesigned workflows, and three layers for people and AI agents to work together.
06 · The HR mandate
The goal is not to monitor every prompt. It is to make consequential work legible enough for people to learn, challenge, decide, and ask for help.
Define when material AI contribution is disclosed in the work—not as a confession, but as part of authorship and handoff.
Name what a person must read, challenge, and verify before an artifact can move forward.
Set decision rights, stop conditions, escalation paths, and the person accountable for release.
Teach healthy reliance, preserve the right to disconnect, and give managers a non-diagnostic path for responding to observable concerns.
Inspect the work. Support the person. Never turn private prompts into a performance score.
07 · Start here
Use your current policy and three synthetic scenarios. No new tool, employee data, or production change.
Your current policy—not this page—supplies the answer. If the policy is silent, record an escalation question.
A manager uses the company-approved AI assistant to draft an FAQ from HR guidance that is already approved. Before publishing, the manager compares every answer with the source.
An AI-drafted manager brief adds a recommendation that does not appear in the approved source material. The manager notices it before sharing.
A recurring AI-produced manager update is ready to send, but the current guidance does not say who must verify the claims or approve release.
45-minute agenda
Copy these columns into a sheet
Found disagreement or a missing owner? Bring the gap list to a 20-minute working session. Leave with a one-page map. No deck.
This is a learning and policy-comprehension exercise—not a policy change, employment decision, employee assessment, or AI deployment. Follow your organization's current approval and acceptable-use rules.
08 · If the check finds a gap
Once the accountable owner has approved any needed clarification, test the guidance with one willing group for two weeks under your existing rules. This is still an HR-led learning test—not the cross-functional operating sprint below. Done means four artifacts exist.
Every question the guidance did not answer, written down as the manager asked it, with the repeats marked. For example: What must a person verify? Who decides when the policy is silent? When must AI contribution be disclosed?
The questions your approved policy does not answer yet. This is the artifact that earns the conversation with the business owner.
The same practical questions, asked of the same managers before and after. You can see whether the answers converged.
The route from “can I use AI here?” to a decision, plus the escalation they took when the path ran out.
If the two weeks end with an empty question log, stop and find out why before doing anything larger. Either the managers never used the guidance, or nothing in their work was ambiguous. Neither one is a reason to roll anything out.
Our approach
Today, people divide the brief, work with AI agents in private, and merge plausible sections at the end. The executive meeting becomes the first real integration step.
Private contexts → plausible sections → conflict at executive review
One shared brief → visible contributions → named human review → resolved decisions
Decision, evidence, roles, the review standard, and what counts as done enter the room together.
People and their AI agents work by role inside one traceable process.
Each reviewer reads the integrated artifact and challenges the claims in their area.
AI agents can surface the conflict. It stays open until the person with decision authority resolves it.
How 8Hats Lab helps
We test these three layers in real work and hand the company the ones that hold up.
Personal AI agent
With explicit consent, AI agents can learn how a person prefers to be challenged, how ambiguity is escalated, and when control returns. Personalization is support—not employee scoring.
Rooms
Shared briefs, roles, contributions, review gates, open conflicts, escalation, and an agreed definition of done.
Explore Agents CoworkingOrganization Model
Accepted decisions, corrections, and working standards live in the company's shared knowledge instead of private chats.
See the Organization ModelThe handoff
If the answer requires a new tool, data access, a policy change, an employment decision, or a change to how a business team works, bring in the accountable owner. HR can surface the gap; it should not make those decisions alone.
We can map that handoff with HR, the business owner, and the people who do the work. Technology, Security, Privacy, or Legal joins when the identified tool, data, policy, or decision boundary requires it.
We start with the next operating result. If the first case does not earn the second, there is no second.
Cross-functional next step · two-week operating sprint
Who you would be working with
Our team has led learning, research, product, and transformation from inside universities, corporations, public institutions, and companies of our own.
For 25+ years, he has put digital tools, and now AI, to work in learning and education. He has seen it from every side: as an educator and researcher, a corporate manager, and the owner paying for the result.
LinkedIn
For 25+ years, he has built learning at scale. His courses and platforms have reached 7M+ people, with dozens of companies and universities as partners. For the past two years, he has designed AI systems that capture both documented and tacit organizational knowledge.
LinkedIn
An engineer by training, with a PhD in dynamics and strength of machines. He spent 20+ years inside universities as a researcher and manager, including as vice rector for academic affairs at a technological university. He knows how a large institution actually changes the way it teaches.
LinkedIn
A PhD economist with 14 years across data, ML, NLP, and product in banks, central banking, and fintech. She works with leaders on the questions that decide whether AI agents help: what to automate, what to measure, what stays human, and who holds the decision.
LinkedInFor search—and the first internal conversation
A human–AI team is a working unit in which people and AI agents contribute to the same outcome. AI agents may draft, research, coordinate, or challenge; people retain the decision rights and accountability defined by the organization.
Because AI changes roles, contribution, review, learning, trust, management practice, and employee support. Technology, Security, and Legal remain essential owners of systems, controls, and policy. HR owns the human and organizational design with them.
Shadow AI is employee use of AI tools or AI agent capabilities outside the organization's approved systems and controls. An inventory helps, but the operating question remains: what must be disclosed, reviewed, escalated, and approved when AI contributes to consequential work?
Set evidence and review standards around consequential artifacts, train people to challenge fluent output, define stop and escalation conditions, and offer support through established wellbeing channels. Do not inspect private prompts or use inferred mental states as performance data.
No. The 45-minute check uses your current policy and synthetic scenarios. It does not require an AI tool, employee or candidate records, individual scoring, or access to anyone's prompt history. Follow your organization's current rules for any later activity.
Download the check and run it first. If managers disagree or the policy runs out, use the gap list to choose one recurring workflow and map its people, AI agents, evidence, handoffs, review points, and decisions.
A bounded first move
Use the guide to name the working version, contributors, reviewers, review gates, and decision owner. Then bring the first unresolved gap to a 20-minute working session and leave with a one-page workflow map.
Bring one document. Keep the map. No deck.
PDF · 7 pages · roles, workflow, review gates, and a one-document checklist
