AI was supposed to hand HR its time back. Free up the people team from the busywork so they could finally do the human work: coaching managers, designing culture, sitting with people through hard moments. For a lot of teams, it has done the opposite, at least for now. The tools multiplied. The dashboards multiplied. And the people work kept waiting in line behind all the building.

That tension framed the final session of our Rebuilding the Modern Workplace series, a conversation about what it actually looks like to design a workplace instead of just managing one. Rebecca Taylor hosted, joined by three people leaders who are living this in real time: Brandon Sammut, Chief People and AI Transformation Officer at Zapier; Josh Greenwald, SVP and Chief People Officer at Sword Health; and Claire Schmidt, founder of AllVoices.

What follows is a full breakdown of the conversation: the frameworks, the disagreements, the spicy takes, and the specific things these leaders are building and scrapping right now. If you missed it live, this is the whole thing.

What It Means to Design a Workplace, Not Just Write a Policy

The opening question cut straight to the theme. What does it actually look like to design how work works, rather than writing one more policy and calling it a program? Brandon Sammut reframed the whole exercise as a design discipline, the same one product and security teams already use.

"The best design work I've ever seen starts with a really clear statement of the thing that we're solving for. It's a problems-based effort, not a solutions-based effort."
— Brandon Sammut, Zapier

His point: most people quietly flip the head and the tail. They lead with a solution and reverse-engineer a problem to justify it. Good workplace design starts with a problem or opportunity worth solving, then generates one or more ways to solve it. And the ratio of time matters. For every hundred minutes a people team spends, how many go to understanding the problem versus jumping to solutions? The answer reveals how much the team actually values solving the right thing.

"We can be exceptional problem solvers. But if we're not solving the most important things, we're missing most of the game."
— Brandon Sammut, Zapier

That distinction lands harder than it sounds. HR teams are wired to surface problems; every function is. Sales says we're not selling enough. Marketing says we're not marketing enough. The harder discipline is sequencing: which problems are the ones this team is uniquely accountable for solving, and in what order, so you stop playing organizational whack-a-mole. It is a pattern researchers keep finding: layering AI onto legacy workflows produces incremental gains at best, while the real value comes from reimagining how the work itself gets done.

How do you design a workplace in a highly regulated industry?

Josh Greenwald runs people strategy at an AI-native healthcare company, which adds a regulatory layer most teams do not carry. His answer was that the principles do not change much. You can be wildly innovative inside a regulated environment as long as you have the operational discipline to match.

"We don't experiment with patient safety or clinical outcomes. Those things are sacred."
— Josh Greenwald, Sword Health

At Sword, the nonnegotiables stay nonnegotiable: a human always in the loop, documentation and auditability, clear escalation paths, strong governance. Inside those guardrails, the team experiments aggressively with workflows, intelligence, and engagement models. The starting point is always the healthcare problem, not the technology. Sword was founded because access to high-quality care was shrinking while costs climbed, all against a clinician shortage. So the question became: what can AI do that we could not do before? The answer looks like a human clinician intake, an empathetic AI care specialist with long-term clinical memory and wearable integration, computer vision guiding a patient through physical therapy, and a human looping back at the end.

"It's really about human transformation. How are we using AI to solve our problems?"
— Josh Greenwald, Sword Health

How to Tell If a Workplace Is Actually Being Redesigned or Just Repackaged

Claire Schmidt has a simple test for whether a redesign is real: are you actually hearing from employees, and are you using what they tell you? She built an entire company on the conviction that the signal is already inside your organization if you bother to collect it.

"I will beat this drum all day every day. Get feedback from employees, and use that feedback to actually design improvements to your workplace."
— Claire Schmidt, AllVoices

The companies that separate good from great, she argued, genuinely want insight into the employee experience and gather it from multiple sources at once: a feedback platform, an employee relations system, surveys, stay interviews. The more comprehensively you collect it, the better-informed your design gets. This is the difference between continuous feedback loops and a once-a-year engagement survey that tells you what already happened.

What does it look like when employees know the answer and leadership misses it?

Claire told a story that stuck. Years ago she set out to launch a new company benefit. She researched it hard, talked to other founders, built a shortlist of the top six options, and surveyed the company to see which one people wanted most.

"The winner was a write-in. It was not even something I had come up with, with all the effort in the world and all the research."
— Claire Schmidt, AllVoices

The employees knew the answer the whole time. The only thing standing between the company and the right call was the willingness to ask and listen. Which brought the panel to the elephant in the room: the Bolt CEO who publicly suggested the company got better after gutting HR. Claire was unsparing.

"It's almost like an Onion article. All the company's problems went away as soon as we got rid of HR. I bet if you ask any employee, they'll say that is not true. You're just not hearing about it."
— Claire Schmidt, AllVoices

Rebecca's analogy was perfect: it's like pulling the batteries out of the smoke detector and declaring the fire problem solved because the beeping stopped. Cutting the function that surfaces problems does not make the problems disappear. It just makes you blind to them. That is exactly why building a culture of listening matters more, not less, as companies move faster.

What's Actually Working With AI in HR Versus Expensive Theater

Rebecca asked the question everyone is circling on LinkedIn right now: what AI use cases are genuinely working, and what is just expensive theater dressed up to look modern? Brandon's answer started somewhere unexpected, and it built on a theme the series has returned to before, namely where HR teams can start using AI right now. The bottleneck is not the technology anymore.

"If you'd asked me a year ago, my answer would have been very different. There are relatively few things AI cannot do now. But just because it can do something doesn't mean it should."
— Brandon Sammut, Zapier

The secret ingredient for deciding where to apply AI is clarity. If a people team lacks a sharp, specific understanding of what the organization needs in this exact moment, no amount of AI will help it prioritize. Brandon owns that clarity for his team. For context on how seriously Zapier means this, 97% of its team now uses AI in daily work, and every job candidate is evaluated for AI fluency.

How is Zapier using AI to lead organizational change?

Brandon gave the most concrete example of the session, and it is worth walking through because it shows what "AI frees up human time" actually means in practice rather than as a slogan.

Zapier is building fast, which means a high rate of org and role change. Normally a manager planning a change starts with their people business partner, brilliant but scarce, who becomes a bottleneck. The PBP has to take several meetings just to build context, ask sharpening questions, shape a blueprint, then produce the collateral: timeline, talking points, FAQs, the message to the team, the HRIS updates. A whole choreography.

Now the manager starts in an AI tool preloaded with skills, standardized ways of approaching a topic, built by the people business partners themselves. The agent interviews the manager, sharpens their thinking, and outputs a prep document that becomes the agenda for the manager's first real meeting with the human PBP. Three or four meetings collapse into one. By the time the human conversation happens, drafts of all the collateral already exist, and the PBP is aligned and warmed up.

"It sounds hand-wavy when people say AI makes more time for the things that are uniquely human. But practically speaking, it's there to be found if you're looking for it."
— Brandon Sammut, Zapier

The payoff is not just speed. It is that the people business partner gets to spend the recovered time on the one thing that always gets shoved aside: checking in on the manager as a human. How are you feeling about this change? What are you unsure about? Brandon gave the entire credit for the idea to his PBP team, which is the point. The craft of HR did not get automated away. It got extended.

Where does AI beat humans in people work today?

Josh pointed to pattern recognition. AI can synthesize many disparate data points in ways a human brain simply cannot hold at once, and that produces insight. He is bullish on AI coaching specifically, with the same caveat that governs human coaching: it only works to the degree you are willing to open up and connect it to real signal.

"Generally speaking, the more access you give it, the better the insights you're going to get."
— Josh Greenwald, Sword Health

Connect it to your Slack so it can interpret your messages. Let it sit in on some meetings. The more context, the sharper the feedback, the same way a credit score only helps once you know it and pair it with an action plan. Self-perception, Josh noted, is often very different from how others actually experience you. Useful, if humbling.

Will AI Make Better Talent Decisions Than Humans?

This is where the conversation got genuinely spicy, and where the panel did not fully agree. Brandon offered a take he flagged as uncomfortable on purpose.

"It may well be the case in the future that, for certain types of talent decisions, AI will make better decisions. Whether better means more evidence-based, more consistent, fairer, or more timely."
— Brandon Sammut, Zapier

His caution to his own team is not "never let AI make a talent decision." It is to hold an open mind that for some decisions, a future model paired with humans rotating into the design and auditing of those systems might outperform humans making the calls directly. He named the obvious constraint himself: from a regulatory standpoint it may simply not be allowable. A growing set of state laws now holds employers accountable for AI-driven employment decisions, and most require a person to stay in the loop before an automated call affects someone's job.

But Brandon's deeper point was about trust, not accuracy. A decision can be technically more correct and still land badly.

"You can technically be right about a thing and still get it wrong. If the decisions are slightly more correct but the impact is incrementally negative, I'm not sure what we're doing here."
— Brandon Sammut, Zapier

How does AllVoices think about AI making decisions about people?

Claire held firm on human-in-the-loop as the line AllVoices does not cross. Every decision gets signed off by a human. But she made a sharp distinction that reframes the whole debate.

"Everything has to be ultimately signed off on by a human. But that doesn't take away from the 99% of the effort to get to that decision point. AI can do a lot of that."
— Claire Schmidt, AllVoices

She used a live example: a new AI-enabled leave of absence workflow her team is designing. Getting a leave approved involves a long chain of near-identical steps, paperwork, and boxes to check. That is a poor use of a person's time and a strong use case for AI. At the end sits a decision point, and Claire conceded AI might be better suited to draft it: here is the recommendation, here is the data, here is the precedent, here is your policy. A human still presses yes. The condition is non-negotiable: the AI has to justify how it reached the recommendation. "Because of all the cases I've looked at" does not cut it.

The chat sharpened the framing further. As one attendee put it, AI does not automatically make better decisions; it often creates better conditions and considerations for better decisions. The panel embraced that, with everyone reserving the right to get smarter as the tools and the evidence evolve. This is the philosophy baked into the Vera AI agent for employee relations: do the heavy synthesis, surface the recommendation with its reasoning, and keep the human holding the pen.

How do you handle attorney-client privilege when AI sits in on sensitive conversations?

An audience member raised a genuinely thorny one: many reorgs are attorney-client privileged, so bringing an AI agent into the fold gets hairy when you think about what becomes discoverable. Brandon did not dodge it. AI notetaking is discoverable, full stop. His team applies a simple ROI calculus: if a use case introduces brand-new legal risk for marginal benefit, maybe that one waits. But in the org-change example, the speed and quality gains were large enough that the incremental privilege risk felt modest and manageable. The deeper safeguard, he noted, is that the actions and decisions you can defend matter more than any single artifact.

What HR Should Stop Doing Right Now

Rebecca closed the main conversation with the spiciest prompt of all: what is HR wasting its time on that nobody has the guts to call out? Josh went first, and he went big.

"Almost all traditional forms of L&D. I have no interest in doing most forms of traditional L&D at this point."
— Josh Greenwald, Sword Health

His reasoning is worth sitting with. Think about how people actually learn a trade: welding, plumbing, pipe fitting. You apprentice under someone who has done it for years, in the flow of the work, getting feedback boom-boom-boom while you do the thing. The same is true in sports. The clock speed of feedback is incredibly high because the coach is right there. People grow fast in that environment.

Knowledge-worker L&D is the opposite: classroom or e-learning sessions outside the context of real work, with little chance to practice and almost no live coaching. For thirty years people dreamed of "learning in the flow of work" and the technology was not there. Now it is, which is why Josh has lost patience with the old model and leans toward upskilling that future-proofs the organization through real practice instead of classroom hours.

What else is HR wasting time on?

Josh kept the list coming. Traditional workforce planning, which he reframed as simply "work planning": give people budgets, use tools to manage them, and stop burning quarters on headcount reconciliation and req approvals, a shift that pairs naturally with measuring output instead of hours worked. Comp decisioning can be sped up dramatically. Offer management, one hundred percent. And leave of absence administration, which he flagged as a current pain point at Sword and an open invitation to compare notes with Claire.

He added an aside that landed with the audience: his two high-schoolers are discouraged from using AI in school, and he thinks that is the wrong call. The skill we should be building, in school and at work, is using AI to become a better problem solver and a more curious thinker, not pretending it does not exist.

How to Build AI Coaching That Challenges People Instead of Flattering Them

One of the strongest audience questions of the session: how do you design AI coaching infrastructure that stress-tests assumptions and challenges managers, rather than just reinforcing what they already believe? Brandon lit up, because his team has a specific method for exactly this.

"Our LLMs tend to be a little sycophantic. They really want to please. They don't always want to challenge. And we need to be challenged if we're going to get better."
— Brandon Sammut, Zapier

He shared an observation a lot of people are quietly noticing. People absorb hard feedback from AI more readily than the same feedback from a human, a dynamic that reshapes how teams develop more impactful managers.

"Teammates can take very critical, sometimes biting feedback from AI and do something with it in a way that feels much lower stakes and less painful than getting that same feedback from a person."
— Brandon Sammut, Zapier

The mechanism Zapier uses to detune the flattery is what they call a rule, distinct from a skill. A skill standardizes how you approach a task. A rule is closer to a hard-coded way of working that applies to everyone in a given AI tool. Some of those rules deliberately bake in the company's timeless principles around metacognition, thinking about how you think, so that critical, first-principles reasoning shows up consistently everywhere people work with AI. They used to train these mindsets in onboarding. Now they productize them.

Josh seconded it hard. Sword built a persona around what makes someone successful there, almost an ethnographic study of the culture, and embedded "be first-principled" into how their AI challenges assumptions. The shared lesson: if you can encode a cultural operating principle into your LLM, you get consistency across everyone's interactions, which is exactly what you want when you are scaling AI skills across a company. None of this works without a foundation of psychological safety, the thing that lets people actually act on hard feedback instead of getting defensive.

The Throughline: Stop Searching for the Right Playbook

If there was one idea connecting every thread, it was this. The companies that win the next decade will not have better policies than everyone else. They will have better feedback loops, better judgment about which problems to solve, and the discipline to design their own systems instead of copying someone else's context, which often means unlearning and redesigning how work happens rather than tweaking it. That is the same argument the panel built across the whole series, from the kickoff on what's broken to session three on building internal intelligence.

Brandon's closing thought captured the mood: these questions are timeless, but they feel more timely than usual right now, and we are all better off having forums to keep pulling on the threads together. Josh framed the arc cleanly: what problems do we bring forward, how do we prioritize solving them, where do we reinvest the time AI gives back, and what do we finally stop doing?

For the people teams asking those questions, the infrastructure to act on the answers matters as much as the answers themselves. That is the work the AllVoices HR case management platform is built for: turning employee signal into decisions you can actually defend.

FAQs from this discussion

Want to see how AllVoices turns employee feedback into decisions you can defend? Book a demo with the AllVoices team.

‍