The Hard Part Comes After AI Works
![[OPT 1] IMG PMP – Editorial – Dinner Conversation with CHROs-14152](https://peoplemanagingpeople.com/cdn-cgi/image/format=auto,quality=82,onerror=redirect,width=792,height=446,fit=crop/wp-content/uploads/sites/3/2026/09/OPT-1-IMG-PMP-–-Editorial-–-Dinner-Conversation-with-CHROs-14152.jpg)
Start With Work: Companies should redesign workflows around business problems before selecting AI tools or measuring expected returns.
Use-Case Judgment: AI training creates more value when employees learn to identify worthwhile problems, not merely operate new tools.
Job Redesign: Successful AI adoption forces leaders to redefine responsibilities, skills, performance standards and career paths as tasks change.
Scale Carefully: Pilots rarely transfer directly across functions; organizations must standardize proven practices while adapting for workflows, risks and data.
Fair Access: Unequal AI access can distort performance judgments, development opportunities and productivity comparisons between employees doing similar work.
Last week, I sat down with a group of CHROs in Atlanta to talk about scaling AI across the business.
We started with ROI, but the conversation kept returning to the work itself: which problems deserve AI investment, what organizations should do with the capacity it creates and when enough tasks have changed that a job needs to change with them.
Kia Painter gave us our starting point. Begin with the business problem, understand the workflow, redesign the process and then decide where AI belongs.
Continue Reading for Free
Create a free account to finish this article, plus get ongoing access to timely insights and practical resources.
Have an account? Log In
Painter recently retired from Cox Communications after 28 years with the company, most recently serving as chief people officer. Her sequence matters because organizations have plenty of ways to use AI. The harder discipline is deciding which uses deserve time, money and organizational change.
Use-Case Judgment May Matter More Than Tool Training
A few weeks before the dinner, I ran an informal poll in our AI Signal newsletter asking why managers were requesting more AI training for their teams. The top reason: People weren’t sure which use cases mattered.
That points to a problem with the way many companies approach AI enablement. Teaching people how to prompt, summarize documents or generate content builds familiarity with the technology. It does little to help them decide where AI can improve a meaningful business outcome.
Angela Cheng-Cimini, chief human resources officer at The Chronicle of Philanthropy, described an approach built around employee problems rather than tool capabilities. She said her organization planned to crowdsource day-to-day challenges from staff, choose several that had broad relevance and run short hackathons to solve them. One employee had already taken that approach herself, building an agent to handle repetitive invoice requests that had been flooding a customer service inbox.
That is a much more useful model for AI training because it connects learning to the work people already understand. The capability companies need to develop is use-case judgment: Can a manager look at a workflow, identify where time or quality is being lost and recognize where AI could make a difference?
That judgment also improves the odds of producing ROI. A technically successful deployment aimed at a low-value problem is still a poor investment.
Capacity Forces Decisions About the Work
The next question is what to do when the use case succeeds.
AI can remove administrative work, reduce cycle times and give people capacity they did not have before. Leaders then have to decide what that capacity is for.
Around the table, that led to discussion about job compression and broader roles. Kesi Dorner, AGM of Human Resources at Metro Atlanta Regional Transit Authority (MARTA) talked about combining work that currently sits across specialized HR positions into more rounded business-partner roles.
My question of when has a job changed enough that it should be redesigned altogether went unanswered.
Those choices are where hours saved turn into an operating decision. A company can use the capacity to increase output, improve service, reduce future hiring, expand responsibilities or tackle work that had been crowded out. Each choice produces a different kind of value.
It also changes what the employee is responsible for.
That became even clearer when the discussion turned to headcount. Painter described conversations with her CFO about whether AI could reduce labor costs. Her response was to go back to the work.
Take accounts payable. What are people actually doing? Which tasks can AI absorb? What remains? What business outcome still has to be delivered?
Her conclusion was that AI needs to be examined through the operating model and the design of the work before leaders start making workforce assumptions.
That is a valuable role for HR because AI often changes jobs incrementally. Where one task disappears, another becomes faster. Routine work may shrink, but the remaining work requires more judgment. New responsibilities may then fill the space that was created.
At some point, the accumulated changes produce a different job.
HR should be watching for that threshold. Job descriptions, skill requirements, compensation, career paths and performance expectations all depend on an accurate understanding of what the role now requires. If leaders keep layering AI onto existing roles without revisiting those fundamentals, employees can end up doing different work under an old job design.
The same issue applies to team performance expectations. If one team becomes more productive with AI, leaders will naturally ask whether the new output level should become the standard.
That decision requires more than proof that the technology worked. It requires confidence that the gain is sustainable, that employees have comparable access to the tools and that the work is similar enough to apply the same expectation across the organization.
Scaling Beyond the Pilot
This is where a successful pilot becomes an enterprise challenge.
A use case can work well in one team because the workflow is structured, the data is clean, the manager understands the technology or the employees have had time to experiment. Another function may have none of those advantages.
Painter described working with her organization's head of AI on a target-state roadmap that connected technology decisions to the future operating model.
The company also created a cross-functional AI council involving HR, technology, finance and legal. The aim was to coordinate investment and governance around business priorities instead of allowing every function to spin up its own collection of tools and projects.
This is a vital part of scaling because you cannot simply copy what worked somewhere else. Leaders need to identify which parts of a successful approach can be standardized and which depend on the work, workforce or risk profile of a particular function.
That should also temper broad productivity assumptions. One strong pilot can provide evidence about a use case. It does not automatically provide evidence about an entire workforce.
Tool Access Is Part of the Performance Equation
Scaling also creates a fairness problem.
Attendees described organizations where employees have access to different AI tools depending on function, budget or policy. Some employees supplement what their employer provides with paid tools they use at home or on personal devices. Others do not have that option.
Once AI affects the amount or quality of work someone can produce, those differences matter to performance management.
Kara Miller, former vice president of talent, learning and workforce transformation at Visa, raised another consideration: If two employees are held to the same output standard but one has access to a more capable tool, how much of the performance difference comes from individual capability, and how much comes from the environment the organization created around them?
The same issue can affect development and career mobility if some roles give employees far more opportunity to build AI skills than others.
Companies do not need identical tools for every employee, the group agreed. They do, however, need to understand when unequal access starts affecting the standards they use to judge people.
A Better Measure of AI Maturity
Near the end of the evening, I asked what would need to change over the next year for leaders to feel confident that their organizations had figured out how to scale AI.
The answers centered on better judgment across the organization: executives who could discuss the technology with some fluency, employees who understood responsible use, managers who could identify worthwhile applications and clearer decisions about access.
That gives HR leaders a useful way to think about the next phase of AI adoption.
If you’re going to get more value from AI, you will have to become better at choosing the right problems to address, redesigning work when those problems are solved and deciding which gains can be reproduced elsewhere.
HR will also have to be more deliberate about what new capacity means for jobs and performance expectations.
