If you are implementing AI at any level in your organization—and at this point nearly every HR leader is—a structural shift in how AI is priced is about to hit your budget, your vendor contracts and your workforce strategy. Most CHROs haven’t seen it coming. This piece is about why it matters and what to do before the bill arrives.
Here’s what triggered it: Anthropic recently restructured its enterprise pricing so that token consumption, the underlying unit of AI compute, is now billed separately on top of a base seat fee. The bundled allowances that made enterprise AI feel affordable are gone. Every major AI lab is watching and will follow. As OpenAI’s Nick Turley, head of ChatGPT, acknowledged publicly: “It’s possible that in the current era, having an unlimited plan is like having an unlimited electricity plan. It just doesn’t make sense.”
This isn’t just one company’s pricing decision. It’s the moment the AI subsidy ends. Your vendors are built on top of these frontier models. When Anthropic, OpenAI and Google reprice, that cost moves through your platform vendors’ P&Ls and lands in your next renewal conversation. The average enterprise AI budget has already grown from $1.2 million per year in 2024 to $7 million in 2026, and 65% of IT leaders report unexpected charges from consumption-based AI pricing, with actual costs frequently exceeding estimates by 30% to 50%. This isn’t a technology problem. It’s a budget problem. And budget problems belong to the CHRO and the CFO together. I wrote about this shift from a vendor and investor perspective in WorkTech, but the implications for HR leaders are urgent enough to warrant a conversation of their own.
See also: How legacy companies can become AI-enabled: from bolt-on to built-in
Your HCM platform is already an agent network, whether you know it or not
The transition to agentic AI in your HR stack isn’t coming. It’s already happened.
ADP, Workday and SAP SuccessFactors are all building agentic AI natively, not as optional add-ons, but as core platform architecture. Forty-eight percent of large businesses already report using agentic AI. SAP SuccessFactors’ 1H 2026 release expanded a connected network of agents across recruiting, payroll, workforce administration, performance and talent development, working together across the entire HR lifecycle without waiting to be asked. Workday is moving from system of record to platform of agents. Oracle has its Agent Studio. Every one of these platforms is simultaneously opening up to integrate with external agents from Microsoft Copilot to Anthropic’s Claude to Google’s Gemini, creating agent-to-agent connections that multiply token consumption with every interaction.
A single agentic HR workflow may trigger 10 to 20 LLM calls. Multiply that across your HR lifecycle touchpoints, the employees bringing their own AI agents to work and the agent-to-agent integrations your platform vendor is building on your behalf inside contracts you signed before any of this existed. The question most CHROs haven’t asked yet: What is my organization’s daily token consumption across my HR stack, and who owns that number? Right now, the honest answer for most enterprises is: Nobody knows.
The productivity paradox
The promise of agentic AI in HR is real. IBM’s AskHR handled more than 16 million employee interactions in 2025, a 65% increase year over year, while significantly reducing transaction times. The efficiency gains are documented and defensible. But there’s a paradox hiding inside the productivity story.
AT&T scaled AI across more than 100,000 employees and implemented a multi-agent system that delivered 90% cost savings at the per-unit level, while tripling total token volume to 27 billion tokens per day. The cost per action went down. The total bill went up. That’s the agentic AI economics equation: Efficiency at the unit level does not equal efficiency at the enterprise level when volume scales faster than you planned.
Klarna is the cautionary tale every CHRO should know. Its AI assistant handled 66% of 2.3 million monthly customer service chats and was projected to save $40 million annually. But analysis of Klarna’s IPO filing showed those savings represented just 1.3% of total expenses. By early 2026, the company was rehiring human agents. The ROI math only holds when you account for the full cost of quality, governance and scale, not just the API bill. Translate that directly into HR: Productivity gains from AI agents are real, but so are the hidden costs of compliance monitoring, quality assurance and the human-in-the-loop oversight that keeps agents from making decisions you’ll have to explain to a regulator.
The workforce planning dimension is what most leaders are missing. Agents doing more work does not automatically mean fewer people and lower costs if token consumption scales faster than labor savings. What looks like a headcount reduction opportunity may actually be a cost substitution, trading predictable salary expense for variable, difficult-to-forecast compute expense. Your CFO will have a strong opinion about that trade-off. You should too, before the conversation starts.
The CFO is already in this conversation. Are you?
Deloitte recently published a dedicated guide to AI token economics for CFOs, a signal that finance has entered the building. Seun Salami, CFO of TIAA Nuveen, put it plainly: “Please befriend your CFO.” CHROs need to hear that message and act on it before the bill arrives.
The budget tension is structural. HR leaders want AI-powered tools that deliver on better hiring, stronger retention and smarter workforce decisions. Finance needs predictable costs and defensible ROI. Outcome-based and consumption-based pricing models, in their current form, often deliver neither, creating variable spend with no ceiling and attribution challenges that make it nearly impossible to prove the software caused the outcome.
Here’s where the public versus private vendor dynamic matters, and it’s worth naming directly. Workday, SAP, Oracle and ADP are under earnings pressure to monetize their AI investments. The pricing shift at the frontier AI layer will flow through their P&Ls and into customer contracts on a timeline driven by earnings, not customer readiness. Private vendors have considerably more flexibility to absorb margin pressure, offer creative deal structures and make pricing decisions without a quarterly earnings call forcing the issue. That doesn’t make private vendors better; it makes them differently motivated. Motivation matters when you’re negotiating a multi-year contract in a rapidly shifting pricing environment.
One more dynamic already playing out: Organizations are funding high-water AI spending by cutting point solutions and reducing headcount through attrition. The math feels clean in a spreadsheet. But the timing is brutal. The spend is front-loaded. The savings are back-loaded. And the assumption that agentic AI will deliver promised outcomes is still largely unproven at enterprise scale. The piper has arrived before the band has finished playing.
The enterprise AI ecosystem problem nobody has solved
HR leaders are making AI vendor decisions in a silo, while the CTO does the same with infrastructure, the CDO with data platforms, the CISO with governance and the COO with productivity tools. Nobody is connecting the dots. The consumption cost compounds across every uncoordinated decision.
Gartner’s 2026 CHRO priorities research makes this explicit: While enterprises typically develop a centralized AI strategy, CHROs must also have an HR-focused AI strategy. Those two things are not the same. The result is a fragmented landscape where HR is buying recruiting agents, IT is deploying productivity agents, finance is building reporting agents and operations is running process automation agents, all consuming tokens, all hitting the same frontier model APIs, all generating costs that nobody has mapped to a unified budget.
The disruption risk underneath all of this is no longer theoretical. A16z made the case this week that HCM is the last large enterprise software category without a serious AI-native challenger, and that is about to change. The argument is architectural: Workday cannot become AI-native without starting over, and starting over is the one thing a public, installed-base company cannot do. Every Illuminate feature is an additive overlay on the same forms-and-approvals engine. Flex Credits exist because both sides need it; every enterprise CIO and CFO has “AI investment” as a top-line KPI and needs to show real spend against it, while every legacy vendor needs to show AI revenue on earnings calls. The underlying architecture question goes unanswered.
For CHROs, this is the signal to start piloting AI-native alternatives now, not to replace your HCM tomorrow, but to pressure test the future before your next renewal locks you in for another three to five years. The CHRO who steps into the enterprise AI strategy conversation as a cross-functional stakeholder with a point of view on workforce impact, cost governance and outcome measurement earns a seat at a table that will define how the organization operates for the next decade.
What this requires of you now
This isn’t a checklist. It’s a set of leadership decisions that will determine whether you’re ahead of this shift or behind it.
Own the token question before finance does.
Most HR leaders don’t know their organization’s AI consumption footprint. Audit which vendors are AI-native versus AI-layered, and establish a baseline for your token consumption today versus what it will look like as agentic workflows scale. If you don’t own this number, finance will, and they’ll make decisions about your tools based on cost, not value.
Treat your next vendor renewal as a strategic architecture decision.
The contract you signed two or three years ago was written for per-seat SaaS pricing. Before your next renewal, whether that’s Workday, SAP, Oracle, ADP or any AI-native point solution, demand transparency on consumption models, token volume projections, cost-per-outcome commitments and what happens contractually if usage scales beyond the original estimate. This is not a procurement conversation. It belongs at the CHRO level.
Get into the enterprise AI strategy conversation, now.
If your organization has a centralized AI strategy and HR isn’t at the table, fix that immediately. Capacity planning, headcount strategy, role redesign, governance and the human-in-the-loop decisions that determine where agents operate are fundamentally HR questions. If you’re not in that conversation, someone else is making those calls.
Pilot AI-native alternatives before your next renewal, not after.
A16z is actively seeking to fund the next generation of enterprise HR systems built for agents, not forms and approvals. The commercial path doesn’t require ripping out your current system. Start with a scoped project in an adjacent HR budget line. By the time your renewal comes up, you want informed experience, not a consultant’s recommendation and a vendor’s roadmap slide.
Ask the question most CHROs aren’t asking yet.
The assumption embedded in most AI investment decisions is that what we’re moving toward is more valuable than what we already have. That assumption deserves scrutiny. Before you commit to the next wave of AI-native HR tools, ask your vendors directly: Can you prove the outcome? Who owns the data that measures it? What happens to your pricing if we don’t hit the targets? The vendors who can answer those questions clearly are the partners worth betting on. The ones who can’t are selling you a talk track.
The bill will arrive either way
The CHRO has always sat at the intersection of people, technology and business outcomes. That intersection just got significantly more complex and significantly more expensive.
The cost of intelligence is now a variable that HR leaders have to own, forecast and defend alongside their CFO. The vendors who understand this moment will restructure how they operate, implement and measure success to genuinely align their revenue with your outcomes. The ones who don’t will send you a bill you didn’t see coming and point to the contract when you push back.
The bill is coming. The only question is whether you’re ready for it.
Credit: Source link
