Many leaders treat agentic AI and skills-based work as separate agendas, but they are one transformation: agentic AI needs work to be deconstructed into tasks and skills, and redesigning work at scale needs AI. The same decisions govern both.
In The Conference Board C-Suite Outlook 2026, chief human resources officers (CHROs) cited the “impact of automation, including AI” as their most pressing challenge for 2026, and 43.6% of global C-Suite executives named “AI/technology” an investment priority, outpacing all others. The purpose of that investment is productivity, but workforce readiness––not technology––is the constraint.
That readiness is built through work redesign: how leaders redesign work will determine the scope of the AI dividend—the productivity gains freed up by agents—and who receives it: investors, customers, or employees. The technology cannot decide that allocation; leaders must. Gains in the form of lowered expenses flow to investors; gains delivered through better service, speed, or quality flow to customers; gains achieved as reduced workload flow to employees. A successful transformation delivers dividends to all three.
This executive summary introduces a five-part report series, along with a companion implementation guide, that provides a framework for how CHROs should manage the transformation over the next 18 months, defining the strategic choices required to turn agentic AI and work redesign into measurable workforce and business value.
Trusted Insights for What’s Ahead®
The following insights are the framework this series builds out: five decisions and commitments that govern the transformation. The first two are sequential—pick the outcome, then redesign the work; the remaining three are standing commitments, determined before deployment. Each is developed in depth in the parts noted below.
- Start from the outcome you want and redesign the work backward from it. Few organizations achieve full skills-based adoption, and it is the wrong goal in any case. Organizations creating value pick a desired business outcome, deconstruct work only where it is pivotal, and repeat. Job-based structures remain the right design for much of the work. (Part 1)
- Decide deliberately which work goes to AI alone, which to people working with AI, and which to people alone. The sequence (deconstruct, then allocate) is the decision that matters most because AI amplifies whatever system it inherits. MIT’s much debated The GenAI Divide: State of AI in Business 2025 analysis found that roughly 95% of generative-AI pilots produce no measurable P&L impact. Whatever the exact figure, the bottom line holds: the failures are overwhelmingly organizational, not technical. Deployment itself changes the workflow: handing tasks to agents demands oversight and exception-handling, so the boundaries must be redrawn as the work and the technology evolve. (Part 2)
- Build the data infrastructure, governance, and quality-checking your AI agents require. Generative AI is probabilistic—a wrong answer arrives with the same confidence as the right one—so quality assurance means ongoing sampling, not one-time acceptance testing. And agents are not software you buy once; they’re a workforce you keep paying for. Pricing, monitoring, orchestration, and updating all generate recurring costs. These costs surprise almost everyone: in Mavvrik’s State of AI Cost Governance Report 2025, of 372 enterprises, only 15% forecast their AI costs within 10% of the actual figure. Make sure all AI agents have a named owner, a budget, and an audit trail. (Part 3)
- Measure what people and agents produce together, not how many people you employ. Counting tasks handed off and costs avoided is not proof of value. The value people and agents produce together relative to their full cost––the capability yield––challenges the age-old “cost-per-head” metric. The measures behind it are the redeployment rate—the share of AI-freed hours redirected into more valuable work—and the attribution rate—the share of claimed value that survives five checks that a careful chief financial officer would apply. In our modeled cases, only one-third to one-half of claimed gains survive those checks. (Part 4)
- Decide in advance where the gains will go, then communicate that decision. Whether AI shrinks the workforce, grows its capacity, or does both across the portfolio is a design choice, not a result discovered after. When Ingka Group’s AI assistant took over the work of handling nearly half of routine customer contacts, leaders reskilled 8,500 customer-service workers into a paid design business rather than cutting them. Coinbase made the opposite call, pairing a 14% workforce cut with an explicit AI mandate. Either can be a strategy; not choosing isn’t. Decide the allocation before deployment and say so—people do not readily adopt what threatens them, and unlike most employer promises, this one is visible and verifiable. (Part 5)
Everything that follows in this series is about the distance between deploying agents and redesigning work so that people and agents create value together.
Defining Agents
An AI agent is a system that can pursue a defined goal across multiple steps and take or recommend actions within specified controls. It can use approved tools and data; remember the goal, the steps already taken, and the results; and carry that context forward through the task.
About the Research
Findings draw on 16 primary research sessions conducted February–July 2026, involving more than 75 senior HR, talent, and AI leaders—12 in-depth executive interviews and four focus groups, including an expert panel convened June 25, 2026—plus several sessions from The Conference Board Councils, a three-hour hands-on Member workshop, and a governed AI-workflow simulation platform used by more than 100 Members.
Many leaders treat agentic AI and skills-based work as separate agendas, but they are one transformation: agentic AI needs work to be deconstructed into tasks and skills, and redesigning work at scale needs AI. The same decisions govern both.
In The Conference Board C-Suite Outlook 2026, chief human resources officers (CHROs) cited the “impact of automation, including AI” as their most pressing challenge for 2026, and 43.6% of global C-Suite executives named “AI/technology” an investment priority, outpacing all others. The purpose of that investment is productivity, but workforce readiness––not technology––is the constraint.
That readiness is built through work redesign: how leaders redesign work will determine the scope of the AI dividend—the productivity gains freed up by agents—and who receives it: investors, customers, or employees. The technology cannot decide that allocation; leaders must. Gains in the form of lowered expenses flow to investors; gains delivered through better service, speed, or quality flow to customers; gains achieved as reduced workload flow to employees. A successful transformation delivers dividends to all three.
This executive summary introduces a five-part report series, along with a companion implementation guide, that provides a framework for how CHROs should manage the transformation over the next 18 months, defining the strategic choices required to turn agentic AI and work redesign into measurable workforce and business value.
Trusted Insights for What’s Ahead®
The following insights are the framework this series builds out: five decisions and commitments that govern the transformation. The first two are sequential—pick the outcome, then redesign the work; the remaining three are standing commitments, determined before deployment. Each is developed in depth in the parts noted below.
- Start from the outcome you want and redesign the work backward from it. Few organizations achieve full skills-based adoption, and it is the wrong goal in any case. Organizations creating value pick a desired business outcome, deconstruct work only where it is pivotal, and repeat. Job-based structures remain the right design for much of the work. (Part 1)
- Decide deliberately which work goes to AI alone, which to people working with AI, and which to people alone. The sequence (deconstruct, then allocate) is the decision that matters most because AI amplifies whatever system it inherits. MIT’s much debated The GenAI Divide: State of AI in Business 2025 analysis found that roughly 95% of generative-AI pilots produce no measurable P&L impact. Whatever the exact figure, the bottom line holds: the failures are overwhelmingly organizational, not technical. Deployment itself changes the workflow: handing tasks to agents demands oversight and exception-handling, so the boundaries must be redrawn as the work and the technology evolve. (Part 2)
- Build the data infrastructure, governance, and quality-checking your AI agents require. Generative AI is probabilistic—a wrong answer arrives with the same confidence as the right one—so quality assurance means ongoing sampling, not one-time acceptance testing. And agents are not software you buy once; they’re a workforce you keep paying for. Pricing, monitoring, orchestration, and updating all generate recurring costs. These costs surprise almost everyone: in Mavvrik’s State of AI Cost Governance Report 2025, of 372 enterprises, only 15% forecast their AI costs within 10% of the actual figure. Make sure all AI agents have a named owner, a budget, and an audit trail. (Part 3)
- Measure what people and agents produce together, not how many people you employ. Counting tasks handed off and costs avoided is not proof of value. The value people and agents produce together relative to their full cost––the capability yield––challenges the age-old “cost-per-head” metric. The measures behind it are the redeployment rate—the share of AI-freed hours redirected into more valuable work—and the attribution rate—the share of claimed value that survives five checks that a careful chief financial officer would apply. In our modeled cases, only one-third to one-half of claimed gains survive those checks. (Part 4)
- Decide in advance where the gains will go, then communicate that decision. Whether AI shrinks the workforce, grows its capacity, or does both across the portfolio is a design choice, not a result discovered after. When Ingka Group’s AI assistant took over the work of handling nearly half of routine customer contacts, leaders reskilled 8,500 customer-service workers into a paid design business rather than cutting them. Coinbase made the opposite call, pairing a 14% workforce cut with an explicit AI mandate. Either can be a strategy; not choosing isn’t. Decide the allocation before deployment and say so—people do not readily adopt what threatens them, and unlike most employer promises, this one is visible and verifiable. (Part 5)
Everything that follows in this series is about the distance between deploying agents and redesigning work so that people and agents create value together.
Defining Agents
An AI agent is a system that can pursue a defined goal across multiple steps and take or recommend actions within specified controls. It can use approved tools and data; remember the goal, the steps already taken, and the results; and carry that context forward through the task.
About the Research
Findings draw on 16 primary research sessions conducted February–July 2026, involving more than 75 senior HR, talent, and AI leaders—12 in-depth executive interviews and four focus groups, including an expert panel convened June 25, 2026—plus several sessions from The Conference Board Councils, a three-hour hands-on Member workshop, and a governed AI-workflow simulation platform used by more than 100 Members.
.png)
Email
Linkedin
Facebook
X
Copy Link
