Conference Board sets five-stage AI agent work framework

The Conference Board has published a five-stage framework for companies redesigning work around AI agents. The report argues that organisations need a new approach if AI deployment is to deliver measurable business value.

The research says many companies are moving from testing AI tools to deciding how tasks should be divided between employees and software. It draws on interviews, focus groups and practical research with senior HR, talent and AI leaders.

That question is becoming more urgent as corporate spending on AI rises. In The Conference Board C-Suite Outlook 2026, 43.6% of executives identified AI and technology as an investment priority, more than any other option.

The report outlines a sequence for executives across business, HR, technology and finance. It says leaders should start by defining the business outcome they want to improve, rather than with the availability of a tool.

Only then should companies redesign workflows task by task to decide what AI agents should do, what people should do and where both should work together. The report adds that people should remain accountable for outcomes even when agents carry out work autonomously.

The framework also calls on employers to rethink the human roles that remain after automation changes a process. That includes reviewing the skills, training, job level and possible pay implications tied to the remaining work.

Cost is another central theme. The report says the expense of running an AI agent can extend far beyond software or model fees because organisations must also fund data access, permissions, controls, monitoring, maintenance and human supervision.

In one model cited in the research, direct AI model use accounted for just 9.4% of an agent's recurring monthly cost. The report presents that figure as an illustration of how operating expenses may be distributed once governance and oversight are included.

Measuring results

The report says companies should assess AI projects by whether redesigned work improves quality, speed, cost or other business outcomes. It warns against treating task completion alone as proof of value.

It also argues that time saved by AI should not automatically be counted as a financial gain. Instead, organisations should show how that time was redeployed and what measurable result followed.

To support that approach, the framework separates gains into three categories: cash savings, avoided future costs and higher-value work. Cash savings refer to spending that disappears, such as overtime or contractor costs. Avoided future costs refer to planned hiring or expenditure that no longer needs to happen.

In the report's framing, higher-value work means employee time redirected to tasks that produce a measurable result. The final stage of the framework asks leaders to decide how any verified gains will be used: to reduce costs, absorb more work without equivalent hiring, reinvest elsewhere or share benefits with employees and customers.

The research also raises concerns about workforce development. It says that if AI absorbs routine tasks, companies may remove work that has traditionally helped junior staff build judgement and experience.

In response, the report recommends deliberate substitutes for that learning, including mentoring, job rotations, supervised practice and a gradual increase in task complexity. It presents this as a way to protect early-career talent pipelines while companies adopt more automation.

Another theme is cross-functional responsibility. The report says IT may oversee the technology itself, but HR should help shape changes to roles, skills, staffing, training and change management wherever AI alters how work is done.

Allan Schweyer, Principal Researcher, Human Capital, The Conference Board, said the issue was not only the technical reach of AI agents. "AI agents can complete more and more work, but capability alone doesn't create value. Companies need to know what outcome they're trying to improve, what the agent should-and shouldn't-do, and whether the redesigned process actually performs better. Otherwise, more automation can simply mean doing the wrong work faster," Schweyer said.

Diana Scott, US Human Capital Centre Leader, The Conference Board, said the work should not sit within one department. "Agentic AI is not simply an IT initiative, nor is it solely an HR initiative. Business leaders need to own the outcome, technology needs to provide secure and dependable tools, finance needs to test the economic case, and HR needs to ensure roles, skills, staffing, and employee experience evolve along with the technology," Scott said.

The research was produced in collaboration with SkillsRight, a nonprofit organisation focused on workforce strategy, organisational capability and change management.