Kelley Steven-Weiss, CEO and Co-founder of Novaworks AI
Kelley Steven-Waiss is an award-winning Silicon Valley founder and CEO, public-company board director, former CHRO and AI-driven transformation leader with over 30 years of experience guiding companies through workforce, technology and operating-model change at scale. Kelley successfully founded, venture-backed and scaled the AI-native HR tech startup, Hitch, before its acquisition by ServiceNow in 2022. Kelley served as ServiceNow’s Chief Transformation Officer and is launching her second venture-backed AI-native HR tech startup at Transform. Kelley is an author of two books, a sought-after speaker and thought leader focused on the future of work.
As companies adopt AI agents to handle more workplace tasks, they face questions about how employees will develop expertise, who remains accountable for automated work and what skills the next generation of workers will need. Kelley Steven-Waiss, CEO and Co-founder of Novaworks, discusses the risks of automating foundational work and why companies need to rethink how they develop talent.
What concerns you most about the impact of AI agents on the future workforce?
This is the risk I worry about most. Entry-level work has always been where people learn by doing. If agents take that work, we break the pipeline that produces future experts. Companies should give early-career employees purpose-driven learning paths: let them do the work first, then supervise the agent doing it and pair them with experienced colleagues who can explain why an output is wrong. Expertise must still be built by hands-on practice. The difference now is that you must design that path with intention because it won’t happen by accident.
How can companies preserve employees’ expertise as more tasks become automated?
Design for it on purpose. Rotate people back into doing the work by hand at intervals. Keep some steps human-owned on purpose, not because the agent can’t do them, but because that’s how you keep domain expertise alive. Make reviewing an agent’s work an active task, where people check its reasoning rather than just approving the result. If you automate away every repetition, you automate away the expertise you’ll need on the day the agent is wrong.
What changes when an employee moves from using AI to overseeing an agent?
Training on AI tools teaches someone how to get a good output. Agent training teaches you how to own an outcome you didn’t produce yourself. With a tool, the human is still doing the work and the AI expedites it. With an agent, the agent does the work and the human supervises it, sets its scope and stays accountable for it. That’s a move from operator to manager. Most companies are still running prompt-writing workshops when what they need is management development.
CEO and Co-founder
Kelley Steven-Weiss
Company
Novaworks AI
What should companies do to prepare employees for that responsibility?
Delegation. Most people have never had to break their own work into clear, verifiable steps and hand it off with defined outcomes, limits and points where a human will step in. Managing agents is closer to managing a team than to using software. One person may oversee multiple agents across different workflows. They need to know how to delegate work appropriately, prioritize what requires their attention and set clear boundaries for each agent. You have to say what “done” looks like, what the agent may and may not decide and when it must stop and escalate. If you can’t describe the work, you can’t delegate it, whether it’s to a person or an agent.
How can organizations tell whether employees are ready to manage AI agents effectively?
What works is practice on real workflows: give people an agent, a live process and a deliberately flawed output, and have them find what went wrong. Short, repeated practice inside the actual flow of work beats a one-time course. What doesn’t work on its own is an e-learning module or lunch-and-learn demo. Watching someone else run an agent teaches you nothing about judgment, and judgment is the primary job.
Measure judgment, not completion rates. Do people catch the errors you plant in agent output? Do escalations happen at the right moments, neither too often nor too rarely? Are the quality, speed and rework rates of agent-assisted workflows getting better? The clearest sign is whether employees can explain what their agent does and where its limits are. If they can’t, they aren’t managing it. They’re hoping it works.
When AI agents make decisions on behalf of employees, who ultimately owns the outcome?
A named human, always. The accountability chain should look the same as it does today. Whoever deployed the agent and set its scope owns its outcomes, the same way a manager owns their team’s output. What changes is that companies must write down the rules between AI work and human work: which decisions the agent may make, which it can only recommend and which it must never touch. Tell staff plainly: “You aren’t blamed for every error an agent makes, but you are responsible for the guardrails you set and the reviews you perform.” Clear boundaries are what makes that kind of accountability fair.
