Kelly Trindel and Kasia Chmielinski join the Future of Work podcast to guide leaders to build responsible AI guardrails with critical thinking and trust.
Sydney Scott
Editorial Strategist, AI
Workday
When AI comes up, the room tends to split.
On one side are the doomers, who believe AI is coming for our jobs, our judgment, maybe even our identity. On the other side are the dreamers, who argue AI will erase busywork, unlock speed, and make every team smarter overnight.
Kelly Trindel and Kasia Chmielinski are not buying either story.
In a recent episode of the Future of Work podcast, Trindel, Chief Responsible AI Officer at Workday, sat down with Chmielinski, founder of the Data Nutrition Project, for a clear-eyed conversation about responsible AI, agentic systems, human dignity, and work.
What emerged from the conversation was less a prediction than a call to action for leaders.The future of AI at work is not fixed and leaders have a role in deciding what it becomes.
As Chmielinski put it, “AI is not happening to us. We are in control of what we do, how we buy, how we provision, the guardrails we put around it.”
“AI is not happening to us. We are in control of what we do, how we buy, how we provision, the guardrails we put around it.”
Kasia Chmielinski
Founder, Data Nutrition Project
Report
Responsible AI can sound like an overwhelming concept. It can feel like something meant for lawyers, engineers, or policy experts. But Trindel and Chmielinski make it much more practical.
For Chmielinski, responsible AI is about building technology in a way that takes both people and systems seriously. They call it “fundamentally sociotechnical,” meaning it looks at the social and technical impacts together. To put it plainly: the code matters, but so do the people that surround it.
“I think we need to really shift out of a techno-deterministic mindset and re-center humanity,” Chmielinski said.
Trindel echoed that idea, “The way we define it at Workday is: responsible AI is the alignment of AI systems to two things – human values, and laws and regulations.” That second part matters because both AI developers and deployers are subject to certain compliance requirements. The first part matters because a system can be legally compliant and still feel wrong, unfair, or careless.
This is where AI becomes a leadership issue. AI strategy cannot sit inside IT alone. It needs people who understand risk, work, trust, and impact.
Before scaling AI, leaders need to ask what the system should never do, when humans must step in, who owns the outcome, and how workers will learn to question the tool. These questions do not slow innovation. They make it last.
Chmielinski’s idea of “cognitive surrender” gets at one of the quieter risks of AI at work: people may stop thinking because the tool seems ready to think for them.
Everyone who has used AI has felt this pull. The answer appears quickly. It sounds confident. It is clean, polished, and easy to accept. The brain relaxes and the work feels done. That’s the danger.
Chmielinski warned, “There's a lot of research showing that people just take what the AI says and feel good about it. According to Wharton research, in 80% of cases when the AI is wrong, people will take the answer as true and feel confident. So it's less about what the AI is going to do to us and more about how we preserve and think about our own humanity going forward—retaining our skepticism and staying smart in terms of critical thinking.”
That reminder matters at work. AI can raise the floor. It can help people draft, summarize, search, compare, and move faster. But as Chmielinski noted, it may not raise the ceiling on its own. Experts can spot mistakes. Beginners may not.
That means AI is safest and strongest when it augments expertise rather than replaces it.
A finance expert using AI to scan trends can still question the numbers. A manager using AI to prepare for a hard conversation can still bring empathy and lived experience. In each case, humans don’t step aside. We get sharper.
That may make healthy skepticism one of the defining workplace skills of the AI era.
Not cynicism. Curiosity with a backbone. It means pausing long enough to ask where an answer came from, what might be missing, which assumptions are baked in, what an expert would challenge, and who could be harmed if the recommendation is wrong.
AI can be a powerful partner. It should not become the final voice in the room.
“Responsible AI is the alignment of AI systems to two things – human values, and laws and regulations.”
Kelly Trindel
Chief Responsible AI Officer, Workday
Agentic AI raises the stakes because it can do more than answer. It can act.
Chmielinski describes agentic systems as AI with added access to data, memory, and tools. These systems may understand more about a person over time and may be able to take action on that person’s behalf. A travel agent bot, for example, might search flights, use stored preferences, access payment information, and book a trip.
That can be useful. It can also go sideways fast. Chmielinski puts it plainly: “We decide how much access these systems have and what they do and don’t do on our behalf.”
This is one of the clearest action points for organizations. Before agentic AI gets broad access, leaders need to decide where the gates go. Some actions may be fine for AI to complete alone. Others may need approval. Some may need full human control. The answer won’t be the same for every process.
An agent that schedules meetings may need light oversight. One that approves expenses or triggers customer communication needs much stronger controls. The more impact a system has on people’s money, jobs, rights, safety, or trust, the more careful the organization must be.
Chmielinski also points to the power of defaults. If a system defaults to yes, people may approve without thinking. If it defaults to no, it may slow work or block value. If it escalates to a human, the human needs time, training, and real authority to review the decision. Guardrails aren’t just technical, but cultural.
Boards need a clear AI strategy. Leaders need red lines. Teams need regular review cycles because what works today may not be relevant in six months. Employees need to know when AI is being used and what role they play in checking it. This is the human layer of AI governance, where values become habits.
Report
The conversation also returned to a basic truth that AI hype often skips: bad inputs create bad outputs.
Chmielinski’s Data Nutrition Project focuses on bias and data quality because, as they said, “garbage in, garbage out.” But the point goes beyond having more data. In fact, they argued that the old idea of throwing more data at every problem has run its course. The real need is better data, clearer access rules, stronger documentation, and a better understanding of how data flows through AI systems.
That matters even more as companies buy models, connect systems, and build AI into daily work. If organizations do not know what data a system uses, where it came from, who can access it, and how it is used during training or inference, they cannot fully understand the risk.
Trindel connected this to regulation, including the EU AI Act’s focus on documenting the representativeness of training data for high-risk AI systems. “One of the practical challenges is that if you're purchasing the model from elsewhere, you have to make sure you're somehow gathering that information from the developer or from your use of the system.”
Chmielinski offered a useful way to think about this. AI is becoming a supply chain. Just as companies expect visibility into software components, they will increasingly need visibility into AI components. What went in? Who touched it? What changed? What risks traveled along the way?
Trust also depends on disclosure. When people are interacting with AI, they should know it.
Chmielinski pointed to the harm that comes when people feel tricked, from voice scams that sound like a loved one to services that seem real but are not. The damage can be emotional, financial, or both.
They also raised a quieter risk. If people get used to mistreating systems that seem human, what does that do to how they treat actual humans? Human-like AI is not the problem by itself. It can help people rehearse difficult conversations, practice empathy, or prepare for real interactions. The line is whether people understand what they are dealing with and remain in control. Clarity is a form of respect.
“The two things I would really ask everyone to continue to have — and make sure your teams have — are critical thinking and creativity.”
Kasia Chmielinski
Founder, Data Nutrition Project
The AI conversation is moving fast. New tools arrive every week. New claims fill every feed. New fears and promises compete for attention. But Trindel and Chmielinski’s conversation cuts through the noise.
We do not need to treat AI as magic, and we do not need to see it as destiny. We need to shape it with intention.
That starts by moving away from what Chmielinski called “techno-determinism,” the belief that technology decides our future for us. It doesn’t. People do. Leaders, teams, buyers, builders, and workers all have a role in deciding how AI shows up at work.
The practical work is less glamorous than the headlines. It means choosing tools for real problems, not novelty. It means knowing when a person must own the decision, where the data is weak, how workers are expected to challenge an output, and what guardrails need to exist before a system touches something high-stakes.
It also means leaving room for imagination. AI may change the pace and texture of work, but it should not narrow what people believe is possible. As Chmielinski said, “The two things I would really ask everyone to continue to have—and make sure your teams have—are critical thinking and creativity.”
Do not hand work over to AI. Build with it, question it, and improve it. Keep people at the center, not as a slogan, but as a practice.
A gap between effort and impact is the real ROI problem. Discover the solution.