# Redesigning Work for Human-AI Collaboration

**Podcast:** HBR IdeaCast
**Published:** 2026-09-17

## Transcript

I'm Adi Ignatius.
I'm Alison Beard, and this is the HBR IdeaCast.
So every Thursday for the next month, we will be exploring how the rapid advancement of AI is changing what it means to be an executive.
We'll look beyond the latest technology news and the cost of investment to better understand what AI means for you as a leader, for how you make decisions, and for how the fundamental structure of your organization is changing in lasting and unexpected ways.
First up, today, we're considering the real impact of AI on the culture of organizations and what it means for talent management.
It's about more than having a human in the loop and workforce restructuring.
And here to help us tackle that is Paula Goldman, Salesforce's chief ethical and humane use officer.
She also advises the U.S.
government on AI policies.
She argues that while many organizations are focusing on how AI can make their workers more productive, reduce headcount, and create efficiencies, the deeper promise may depend on redesigning work to fully integrate human and AI talent together.
Today, I'll talk to her about delegating decisions to AI, handling the employee resistance that's out there, and maintaining accountability, all while trying to build trust.
Goldman is the author of Manage the Machine, How to Harness Human-AI Collaboration at Work.
I want to explore the management choices and options that AI is creating.
And, you know, maybe to frame this, you talk in the book about moving from human in the loop to human at the helm.
Talk about that distinction.
If you don't mind, let me back up and say that phrase human in the loop came from the Cold War.
Actually, it came from military when all of a sudden technology could, for example, detect an incoming missile or something like that.
The obvious question was, okay, but who makes the consequential decision about this information?
How do we create a system where people were making the consequential decisions?
That phrase, though, human in the loop got, I think, kind of misunderstood in this wave of AI.
It kind of got framed like, okay, AI is going to draft and people are going to approve.
And that does not really work very well for every single thing when you're talking about AI agents.
The whole point is that they're able to reason through lots of complex tasks and do lots of things all at once at incredible speed.
And so instead, we need a system where we are putting the right things for human judgment at the right time.
That is what we mean when we say human at the helm.
We talk about that as a design principle for AI systems, but I actually think it's also a great metaphor for how we manage in the age of AI.
All right.
So how should leaders decide about, I know there's no general rule on this, but what to give entirely to AI to run, to manage?
What should be done jointly with humans?
What should be fundamentally human?
I mean, you know, obviously it's case by case, but is there kind of a decision rule that can guide executives?
I have to say, just starting by.
asking the question is a very big step forward.
And because I think the early days of this wave has been a lot about just like, get people the licenses, give them a token budget.
And all of a sudden, now we're in this more strategic phase where people are asking.
What actually is AI good at and where does it have flaws and where do we want human judgment to carry the day?
I think that the general stereotype that you hear, the general kind of received wisdom of like AI can handle the routine and people handle the more complex things, like it gets you about 70% of the way there.
But there are a lot of other things to take into account as well.
And that includes customer preferences and emotions.
employee preferences in emotions, questions where there may be sensitive topics that really only people could handle, even if AI can.
And I think the richness is really in kind of thinking about it from a disciplinary perspective, like what does it look like in marketing or sales or service?
Yeah.
And embedded in all this is how we think about AI.
And it's, I don't know if this is metaphorical or real, but, you know, is it software?
Is it a coworker and a teammate?
I mean, how do we really think about that?
Yeah.
Well, actually, I think that's a super fascinating question.
People get very upset when we anthropomorphize AI.
It's somewhere in between.
I like to say AI needs to be managed as a teammate, but not a human one.
It's a collaborator, but it's not a human one.
But there are a lot of skills of management that actually really do apply to AI.
And I argue we are all going to need to know how to manage it because it is a collaborator and it's going to be critical.
We're all managing multiple AI agents in our job or will be.
So that's kind of the central imperative.
As companies look at these options and these potentialities, I don't know where we are in the sort of wave of AI adoption or exuberance or disappointment, but CEOs still feel pressure to show an AI productivity payoff.
Yeah.
I guess the question is, is there a risk or an opportunity cost in treating AI primarily as a cost-cutting technology?
Productivity and efficiency are really important, right?
I don't think anyone would argue with that.
And I don't think anyone would argue that that is a key benefit of AI.
The question is, what happens when you take that too far and you ignore the other goals?
Productivity and efficiency is not the only goal of one's organization or business, right?
So you could think of lots of examples of where if you only take that into account and you don't take the so-called human side into account, you end up with worse business outcomes.
So take customer service.
This is arguably one of the places where AI has the most product market fit.
I don't want to wait on hold for an hour to get an answer about whether I can get a refund for something.
But there's lots of evidence that says when people are angry, they want to talk to a person.
When people are embarrassed, they want to talk to AI.
Some people just want to talk to a person, even if AI could answer a question.
Let's say there's like my friend who works in health care was saying when someone has a new cancer diagnosis, her company makes sure a person schedules that first appointment.
That's not because AI is not capable of it.
It's because there's something to preserve there.
And that's, I think, where the analogy breaks down is that you don't want to lose a customer or have a terrible customer experience because you've extended it too far.
So you said something earlier that we should be aware of anthropomorphizing AI.
But the fact is...
AI is so anthropomorphic.
It is, yeah.
When we deal with ChatGPT, it adopts a kind of overly friendly language.
Yeah.
You know, I've heard people say, what you don't want to do is try to fool people.
When you cross a line and you're trying to fool people or the result is that you have confused people, that's a real no-no.
Do you agree with that?
Is that a risk?
I think that is a risk and it's worth paying attention to.
I think the risk is different for different use cases, right?
So it's a more severe risk when you're talking about AI companions than it is for customer service, but still a risk.
I also think there's other reasons that you want to make sure that you maintain like a little bit of friction.
on a little bit of people understanding that they're managing AI, not a person, because they're different strengths and weaknesses, right?
Why do we have, for example, lawyers being cited for hallucinations and court filings years after ChatGPT came out?
People need to understand that AI can make mistakes, right?
And that's why we build a little bit of friction when there's a decision that really matters where you need someone to take a beat.
and not just sort of, so to speak, cognitively offload the decision.
You want to have a little space for people to actually make sure that they are exercising accountability and oversight.
I'd love to hear you cite one or two examples where AI is allowing companies to achieve more than simply these efficiencies that we've talked about.
And like you, I don't want to minimize the value of efficiencies.
Part of the answer comes in thinking through, well, what do you do with?
the efficiency that you've gained?
And the second part is, how do you, in fact, leverage AI to make the human part of the business stronger as well?
So let me start with the second piece of that.
One of the places I started out really skeptical was the use of AI to help managers manage people better, right?
And this is like a longstanding issue in business, right?
Like, you know, the old aphorism, like people don't leave companies, they leave managers, right?
And if you ask like so many HR professionals that they all told me, the difference is often just like, are these people engaging?
Are they having the hard conversation?
Are they avoiding it?
And I tried all these AI coaches really skeptical, like, it's not going to help me.
And it did.
Like, these are places where you can practice that hard conversation where AI nudge tech is going to tell you, like, your employee survey says that your team wants to be recognized more and they just.
turned in a big deliverable, make sure that you go acknowledge them, go ask a question, et cetera.
These are places where you're tuning the AI to the human side of the equation.
But the first thing I said also was about what do you do with the gains of AI?
How is that part of your...
strategy, right?
So I think a lot about at Salesforce, we are using AI and we're not only producing AI for customer service, we're using it ourselves.
And it's creating incredible efficiency.
And a lot of what we do with that is we think about like, well, what are the new service challenges that our customers are facing?
And what are the skills and needs where we can take the talents of the people that already know our products and know how to serve customers with it?
and redeploy them.
And so we had this huge move to forward deployed engineers, for example.
And this is arguably a kind of turbocharging of that same skill set where they're helping customers use AI in much more powerful ways and not just answer questions, right?
And so there's a multiplier effect.
And I think that's the other piece of it that like we're just starting to see is like, well, it's not just about the AI.
It's like, how do you redesign the workplace around it?
I've heard companies say that if AI can do the entry-level work or the routine work, this allows companies to have this deeper engagement with customers, to be able to create bespoke products and services for customers at a scale that would have been unthinkable.
Does that strike you as the holy grail or one of the potential holy grails for AI in the workplace?
Yeah, and it's not just products and services, it's also experiences.
So think about like...
the AI and marketing, right?
So, you know, personalized marketing is not new.
AI is allowing it to happen at a scale and a speed that is just mind-blowing.
But remember, customers also have AI and they can use AI to filter out some of those same messages, right?
So what is that right balance and how do you create new experiences for the human side of that sort of customer relationship?
And that's what increasingly I'm seeing marketers focus on is, right?
Like, not only how do I make my business and my marketing messages AI legible to the agents that my customers are deploying, but how do we reinvent marketing to stand out in the age of AI?
And I think that's just one metaphor, but it applies, I think, across domains.
I think when people hear the word efficiency, a lot of them think that means reduced workforce.
I'm interested in your perspective on this because I think, you know, the simple answer is, well, you know, we'll cut jobs here, we'll add jobs there.
But more critically, many companies are going to use the power of AI to employ fewer people, right?
I mean, isn't that, don't we have to admit that?
I do not have a crystal ball.
So far, I don't think that has been the case, but I think we have to really take it seriously and we really have to prepare for.
disruption because of AI in the way that jobs and jobs take place.
The focus for me in the book has really been about the other side of the equation that like we really don't talk about.
Like when we talk about AI and the future of work, we're talking about labor market policy, right?
Generally, I'm talking about how do you design how people work with AI?
Because really, when you look at what AI is capable of, it's generally tasks, not entire roles.
for the most part.
And that means that other parts of people's roles are going to become even more important.
How do we make sure that we're like actually getting the right outcomes from AI when people work with it?
You know, this is a familiar question, you know, to both of us, but I'm interested in your take on it.
So if AI, in fact, takes over a lot of entry level work, routine work that younger, less experienced employees, you know, traditionally take on as they learn a profession.
How do you think about the development of the next generation of employees, of experts, of leaders, if that kind of entry-level thing is now being taken over by our AI colleagues?
It's funny, this is the question I get the most.
It's really interesting.
And I think it's because it's real and not, if anything, this is the place where there may be data that's saying, you know, that AI is impacting entry-level work in some domains.
And I guess I'll say a few things.
One, it does not make sense long-term for companies not to have talent that is going to be developed into their more senior roles.
seems to be illogical.
So there's an imperative to reinvent, like, what's the old metaphor?
They worked their way up from the mailroom.
Well, mailrooms don't exist anymore.
We still have the modern equivalent of that.
One of them, I will say from my perspective, is using AI to learn the business.
And I experienced this firsthand because Salesforce made a call earlier this year where we saw possibly companies pulling back from some of that early stage hiring.
And we said, like, this is an amazing opportunity for us.
And we're going to hire a thousand new grads and interns this year.
And I was fortunate enough to have three summer interns on my team.
And I will tell you last week, I sat through their presentations.
We're trying to use AI to solve problems here, right?
And they showed us how to do it better.
Like we were using AI for like prompt injection and they were like, here's a way that you could have it better.
It was incredible.
Like I have never learned so much from interns in my life.
And I think that that is one really important way as we think about redesigning what entry level looks like is managing AI is part of that.
How do organizations need to be redesigned?
How should they redesign themselves now, given what we know about?
you know, AI's capabilities, this sort of, you know, human plus Asian workforce.
I would imagine the design of work is lagging some of these things.
So how do we think about redesigning our offices?
Well, I'll tell you how we think about it at Salesforce, and which is that we have like this really cool division within our HR team that sits down with different organizations in our company.
And is actually using AI to map the tasks that get done to different skill sets and looking at how some of those different tasks are rising as human tasks and some of them are changing and then looking at and redesigning roles of the future.
I may have a biased vantage point on this, but like most of the roles on my team didn't exist a couple of years ago, like a responsible AI architect, for example.
So part of the answer is like really actually creating those roles that are the kind of rising cresting need.
A second part of the answer is actually giving one's own team a seat at the table in that discussion, right?
Not only because there's a lot of uncertainty and sometimes anxiety about what the future looks like, but because People that are closest to the work itself often have like really good insights about where things work or where things are needed and actually even where AI can play a role.
And then I think it's really like using AI to give people a map towards that future.
The kind of extreme version of it, I don't know if you saw the book Flash Teams by Melissa Valentine, but like that I think is becoming possible where people, their skills are legible and we're kind of bringing people together.
changing it as the needs evolve really, really quickly.
We're seeing a slower version of that that is like a more kind of here's where strategically re-architecting around how AI is changing the needs function by function.
And talk about how this actually works in practice.
I talked with a number of HR leaders that are in charge of this sort of internal mobility, this workforce reinvention.
talked to someone at Seagate and I talked to folks at MasterCard and elsewhere, examples of people in government affairs that wanted to learn about security and they use their internal AI talent marketplace and identified like a little gig project that they could use on the security team and then ended up in a role there.
Or people that ended up taking AI skills workshops and classes and participating in internal hackathons and ended up becoming forward deployed engineers.
And there's a lot of technology that's actually quite mature that helps with this.
The interesting part of it, though, I mean, I found all of that really inspiring because these are stories that don't get told very often.
But the interesting part of it actually was the cultural piece that people brought up, which is, you know, we think about bias.
We're used to thinking about it as sort of like demographic bias, right?
We think about bias in AI.
But they were bringing up a different type of bias.
And that was this notion that when people manage their own teams, they're generally looking for people with a particular pedigree, have worked at a particular type of company.
And I think we're in this moment where no one has 10 years of experience with all of these different skills.
And we're going to have to be thinking about a bias towards the future and not the past if we're really going to have the kind of mobility that we want.
But it requires a...
mindset shift, a cultural shift, where people are, you know, actually validating and orienting around these types of skills and open to it in a way that I think has typically been kind of difficult for companies to manage.
What do you mean exactly by a bias toward the future?
If we're talking about creating jobs and roles that have never existed before.
that no one has decades of experience with.
Like, I mean, yes, you can use a proxy for that.
Like, yes, it's probably true that if I'm hiring a responsible AI architect, that like someone that has worked at a big tech company may have relevant experience.
But it's also likely that people are going to come from unexpected backgrounds and that the more material piece of it is like, what are they able to create?
And how do we assess that?
And AI can help us with that, but we have to be asking the right questions first and not just defaulting to these kind of shortcuts.
And so that's, I think, the shift is like the world of work is opening up.
That's exciting.
It's scary.
But we have to be orienting ourselves to an openness to like what these skills really look like versus what we've typically hired for in the past.
accepts the idea that AI will fundamentally reshape their business in the coming years.
You know, what are a couple of organizational decisions that they should make now before it's too late or whatever, before they're in a hole?
Whether it's creating new jobs or creating new departments or creating a new approach to work that we're seeing, you know, as effective in some places that can help people think about sort of planning for this transformation.
I think it's going to be different for Salesforce than it is going to be for a pharma company, for example.
But in all cases, there's like some very clear places where AI is actually, I guess it's kind of a horizontal where I think there are very few knowledge jobs that are not augmented by AI.
And that part we've already seen.
What we're starting to see then is the strategic identification of the places where AI is.
literally changing roles.
So for Salesforce, it's not just customer service.
It's actually like our engineering department is completely transformed by AI, right?
That's like one of the hero use cases of this wave of AI.
And that means every single function that is supporting engineering, including mine, right?
Like we're trying to make sure that all the products that go out the door are trustworthy.
have to then use AI to accelerate all of their processes.
But it's that identification of those new systems that need to be created.
That's going to be different in pharma where AI is not only, you know, you've got the sort of base standard use cases like customer service or marketing or whatnot, but then you've got the AI and science part of it as well.
And I think it's very, very important that CEOs or executives pick a few.
very big bets to focus on in terms of that transformation organizationally, and not just rely on what has been common wisdom these last few years, which is like, give everyone a budget.
Yes, give everyone a budget.
But it is that intentional strategic transformation of these roles.
that makes the biggest difference.
Well, and I think we all blew through that budget.
Exactly, exactly.
But I feel like there are a couple of narratives.
There's a narrative that AI is transforming business in remarkable ways.
It's flawed, but it is doing incredible things.
But another narrative that I think a lot of intelligent experience people have is it produces a lot of slop.
It is frustrating to employees.
And we maybe have overestimated its value, at least in the short term.
Whichever is correct, I do think that sense that AI is producing slop is a thing that exists in your workforce that either has to be proven to be untrue or has to be accommodated in some ways.
I'm really interested in how you think about that, because I view you as essentially a realist, but positive about AI's potential impact.
But there is this, I'd say, very vocal strain of skepticism.
And how do you think about that bounce?
Well, I think there's kind of two questions.
There's sort of the general AI slop question of like, you get a message on Slack and like, did someone write this or did AI write this?
And I actually think our norms are readjusting around that, where it's become, I think, a little bit more acceptable that like, you know, that AI is being used to help with certain work outputs or whatnot.
But the important thing.
And again, this is, I think, the cusp of where we are is that we're really reinforcing that your work product as an individual is your work product.
And you need to take accountability for it.
And AI is very powerful, but it's not a magic bullet.
Like it's not going to solve every single problem.
And so in some cases, we're actually introducing what I was talking about before, like a little bit of friction before you hand this in.
You want to make sure that like you really stand behind every word.
It doesn't matter whether you used AI to do it or not.
I think the other piece of it, though, is really how do you decide actually where not to use AI?
And that's a question we're not talking about a lot.
How do you decide what to reserve for people and why?
And some of that might be like what we talked about before, like the customer preferences and whatnot.
And some of it might be the moments.
that hard management conversation, the time, like your innovation team may want to really go deep on a particular idea before it brings in AI because you're going to get a better outcome.
And there's a whole chapter in the book that's about AI and innovation.
What's the role of AI and innovation?
So IKEA, their innovation team wants to design a new prototype for a couch that breaks all the sort of stereotypes of a boxy, cushiony thing.
And they use AI and it just keeps reverting to the mean.
Why is that?
Because that's generally what AI does if it's not given enough direction.
And so they basically created space for themselves.
It's called front-loading the brief before they gave AI its next set of instructions.
And they started brainstorming things like campfire or gathering space and got really clear on these kind of breakthrough ideas before they gave.
AI new direction to co-ideate with them.
And then they got this prototype called couch in a box, which was this lightweight 10 pound thing that someone could carry around and ended up being exhibited in a museum exhibit in Copenhagen.
So why do I bring up this example?
It's because just defaulting to AI can make for a worse outcome.
for whatever the task is that you're trying to do.
And some of the goal is not to just leverage AI strengths, it's to know where to preserve human judgment or to preserve human creativity.
And that's, I think, the learning cusp that we are on right now in the AI journey.
And that's a big piece of this question around so-called AI slop is it's bringing together the strengths of AI with the strengths of people.
It's about designing human AI collaboration.
So maybe, you know, further on this, you talked a little bit about the front lines and consumer interaction.
This is areas, obviously, where trust is paramount, where you're really, you know, connecting directly with either customer service or more kind of frontline sales.
Do you have any rules of thumb as to where AI is a benefit where you want to be careful?
Yeah.
So we talked about the rules of thumb for customer service in terms of like either the goal of the customer or their emotions, anger, fear or anger, embarrassment and so on.
I think for sales, it's also really interesting, right?
Because, you know, all of a sudden you can use AI to do all the cold calling, right?
Effectively, or you can use AI to.
help all the inbound triaging, all the inquiries that you could never get to before.
And we hear from all of our customers that are using AI for this purpose is like there are thousands of inbound leads that they could never get to.
And so AI can personalize that outreach, right?
Whereas a human was limited to kind of the ones that they perceived as the highest value.
And then it also has its limits, right?
So yes, AI can help make a pitch.
It can.
help understand what the customer is asking about.
It can give a lot of information about the product.
But when you're talking about a complex B2B deal and you're talking about stakeholders within the company that may not be aligned, maybe there was an org reshuffle, maybe someone's under a lot of political pressure, you're actually helping that customer reinterpret what their problem is, right?
And understand it in the context that they're operating in.
And sometimes they won't even reveal that information unless they trust you, right?
It's a relationship question.
And so what we're seeing on these sales teams is that it's just transforming.
There was a study I saw that said like salespeople experience depression at like 3x the rate of normal professionals because they're constantly hearing no, no, no, no, no, no, no, no, no.
Well, Cold calling is probably no longer such a thing, right?
So hopefully the no's are less and it's a more focused, more human experience of sales.
And that's one example of this balance of kind of the human side and the AI side.
Yeah.
What does all this mean for people management?
How it is evolving?
Obviously, we don't know exactly how it will change, but it's already changing pretty dramatically.
So people in the people management business, how do they stay up to speed with?
everything that's happening?
I actually think that people in the people management business are the linchpin for AI, right?
And again, for all the reasons we talked about, the AI transformation is not just a technological transformation.
It's a people transformation because you really, your human talent is still your most valuable resource.
And then how you bring those things together is incredibly important.
I will tell you my hope, and I see all these kind of green shoots of it, is that empowered HR functions use AI to make the people side of things better.
And it's what we already talked about.
It's like the nudges that make managers engage more.
It's the bringing evidence, using AI to bring more data to performance management as opposed to like my recent impression of my employee.
It's even like the AI systems that help really identify people's skills and new opportunities like that new project that they could take on or the new class that they could take on that gives them a bridge to the thing of the future.
I think there's all these super positive ways that HR can leverage AI to transform the company.
And it requires intentionality, right?
Because we all know those stories of AI gone wrong.
People that were otherwise qualified for a job that got screened out.
There are lots of ways it can go wrong.
But leveraged intentionally, it's completely transformative for the human side of the business.
All right, Paula.
Well, thank you for being on HVR IdeaCast.
Thank you so much.
Thanks for having me.
That was Paula Goldman, Salesforce's Chief Ethical and Humane Use Officer and author of Manage the Machine, How to Harness Human-AI Collaboration at Work.
Next time, Allison speaks with Nithin Noria about the biggest surprises new CEOs face.
Plus, on Thursday, we'll present the next episode in our AI series, How the Technology Is and Isn't Changing Communication.
If you found this episode helpful, share it with a colleague and be sure to subscribe and rate IdeaCast in Apple Podcasts, Spotify, or wherever you listen.
If you want to help leaders move the world forward, please consider subscribing to Harvard Business Review.
You'll get access to the HBR mobile app, the weekly exclusive insider newsletter, and unlimited access to HBR online.
Just head to hbr.org slash subscribe.
Thanks to senior producer Mary Dew and senior production editor Kristen Murphy Romano.
And thanks to you.
You're listening to the HBR IdeaCast.
I'm Adi Ignatius.