Playbook for an AI-Native CHRO

Before the playbook, six concepts. They look technical, but they’re the minimum vocabulary for a CHRO to talk as an equal with their CTO and, more importantly, not to be sold to by any vendor.

In Cowork you delegate day-to-day tasks; in Code you build the tools and automations that don’t exist yet.

In the workshop, we ran a live poll, before showing the demos for each level: 335 executives and HR leaders from Brazilian companies answered the question “what AI maturity level is your HR at?”. It’s worth reading the result with this caveat: without yet knowing the real complexity of each level, self-assessment tends to carry a certain optimism.

That distribution matches the global picture. SHRM found that 62% of organizations already use AI in some operation, but only 39% within HR itself.2 McKinsey, in the HR Monitor 2026, concludes that AI has substantial potential in recruiting and employee experience, “but it needs to be integrated into disciplined, well-designed processes, rather than layered on top of existing complex approaches.”4 And the BCG–MIT Sloan study on agentic AI shows that 66% of adopting organizations expect fundamental changes to operating models, roles and careers over the next three years, with HR at the center.5

Translation: being at N1 or N2 today is not a problem: it’s the starting point for the vast majority of the market. The problem is not knowing that the scale exists, and optimizing for the wrong objective function.

The full framework is in our white paper What it means to be AI-Native: 5 levels of organizational maturity in AI adoption,6 built from direct observation of the organizations with the most advanced adoption we know. Here’s the operational version, in the everyday language of HR.

The conventional reading treats the levels as steps, adding incremental capability with each tool. The reality of the most advanced functions is different: each level unlocks a capability that is qualitatively distinct from the previous one, and that new capability multiplies the effect of what already existed. The impact is exponential.

Two implications from the white paper deserve attention, because they run counter to intuition:

Frameworks are abstract. So we took a case that practically every CHRO has already suffered through and ran it across the five levels: closing next month’s payroll preview, requested around the 15th.

The preview is never in one place. Pull the headcount from the HRIS; gather the hires still sitting in the ATS; the terminations in some business partner’s spreadsheet; approved merit raises and promotions that haven’t hit payroll yet. Reconcile it all, compare against the budget, write the narrative for the CFO and suggest an adjustment, if there’s time left. And there are the rules that live in the head of whoever runs it: “Maria’s salary looks lower because she gets R$ 10k of a cost allowance off-cycle; João’s cost is split 50% with another function.” The typical scenario we find: 3 to 5 days to close, every month; 2 to 4 people on the manual reconciliation; zero alternative scenarios.

What changes, in summary

Each level delivers something materially different, not just faster. The payroll preview is only the example: the same pattern holds for every HR job to be done — merit cycle, headcount planning, talent review, recruiting, pay equity, succession, engagement. Each People subsystem crosses the same five levels, and the jump in impact happens in the same place.

The map of the possible

The payroll preview is just one of the use cases, but the impact grows at every level across the entire People portfolio, and each level compounds the previous one.

The most recurring question from the workshop Q&A, and from our clients, is some variation of: “okay, I get the levels; how do I cross from 2 to 3?”. The honest answer: this is the hardest transition on the scale, and it’s not (only) a technology problem. Configuring MCPs is the easy part. The real work is something else.

There’s no AI playbook for a CHRO without this section. Every company has confidential data; HR has twice as much and, in Brazil, under LGPD, people data is among the most regulated there is.

Start with what is not compliant: dumping comp data into a public chat. No guarantee of where the data stays, no per-company isolation, no audit trail, no contractual basis for LGPD. Even with no-training settings enabled, a salary spreadsheet does not circulate in an open cloud. The rules we apply and recommend, by level:

At some point in this journey, someone at the table (sometimes the CHRO themselves, sometimes the CTO) will say: “we’ll build this in-house”. It’s a legitimate reaction. The team is capable, control feels greater and, at first glance, it’s cheaper.

And it’s worth taking the question seriously, because both sides carry real weight. Building in-house gives total ownership of the roadmap and keeps the knowledge in-house. Seeking external help trades part of that control for speed and for accumulated experience that’s hard to reproduce from scratch. There’s also the opportunity cost: every quarter your team spends building infrastructure is a quarter it doesn’t spend doing the People work only it can do. The honest question isn’t whether you can build it — you almost always can. It’s whether it’s worth it, and what you give up to get there.

When we help CHROs work through this decision, we tend to arrive at the same recommendation. And it’s worth a simple mental model to guide the conversation, easy to remember because it’s the argument itself: building with specialized external help is almost always SAFER than building alone. Each letter is a reason.

Building in-house seems cheaper and more under control. In practice, it tends to be the slowest, riskiest and most expensive way to get there. Before opening an “AI engineer for HR” req, talk to someone who has already done this dozens of times.

Where does this converge? Our vision, drawn from what we already operate at our most advanced clients, is an HR function working as a four-layer ecosystem, a single operation, with AI on the inside.

This architecture speaks to what the management literature has been pointing to. BCG–MIT found that 76% of executives already see agentic AI as a coworker, not a tool.5 And McKinsey, in the State of Organizations 2026, argues that the next productivity frontier isn’t in restructurings, but in improving how work flows through the organization.9 The four layers are, in essence, the People workflow rebuilt around an intelligence that connects everything.

There remains the question hanging over all of this, the same one Dan Shipper confronts in After Automation,10 the Every essay that inspired the structure of this article: if AI does more and more, what’s left for humans?

Shipper’s answer, watching a company that automated everything it could, is a paradox: the more you automate, the more expert human work appears: because AI makes yesterday’s competence cheap, cheap competence floods the operation, and the abundance creates demand for judgment: someone to decide what matters now, in this context, at this company.10 Anthropic’s data points in the same direction: the Economic Index shows augmentation (collaborative use, where the human iterates and decides) growing in interactions via Claude.ai.11 And SHRM quantifies the effect: AI’s organizational impact is 5.7× more likely to redistribute responsibilities and 3× more likely to create new roles than to eliminate jobs.2

That’s exactly what the N1→N5 progression describes. At each level, AI absorbs a layer of execution, and the human moves up a layer of judgment. At N1, you still cross-reference data; at N3, you ask the right questions; at N4, you approve, edit and reject plans, and every decision you make teaches the system. The CHRO doesn’t disappear: they become what Shipper calls a framer: the one who defines what’s worth solving, who recognizes the exception the policy didn’t foresee, who knows that promotion can’t be pushed because it’s already been communicated.

“

The future of HR isn’t less human. It’s human somewhere else in the chain: less spreadsheet, more judgment. And it starts with a payroll-preview file and a well-framed question.

The reckoning that opened this article (“what changed because of AI?”) therefore has a three-part answer.

The 30+ skills mentioned are free, open source (MIT) and live at comp.vc/tools/chro-claude-skills and github.com/trycomp-io/comp-skills. The “Claude for CHROs” workshop presentation is at deck-claude-chros.vercel.app, and the recording is on the Comp site.

Theory without an artifact doesn’t change the operation. That’s why, alongside the workshop, we opened up for free the skills package that our HR executives, together with Comp’s product and tech teams, built: the open version of tools that run inside our AI-Native clients.

The skills are model-invoked: you don’t memorize any command, you just describe what you want in plain English (“analyze the pay gap in this spreadsheet by gender and level”) and Claude triggers the right skill. Available at comp.vc/tools/chro-claude-skills and on GitHub.8

Each skill delivers an artifact ready for the meeting (HTML report, analyzed spreadsheet, 1-pager) and runs locally: salary and roster data never leave your machine. On the maturity scale, installing this package is exactly the N1 → N2 move: you leave idiosyncratic use behind and start operating with standardized recipes, with business rules and guardrails built in.

And to build your own skill? Ask Claude itself. It knows the structure (the SKILL.md, the template folders) and creates and installs it for you. Any recurring deliverable that needs to come out the same every time is a candidate.