24 Sep 2026 · 33m · Alison Beard, Adi Ignatius, Wade Foster, Matt Abrahams · 18 quoted facts
AI can make leaders far better informed and faster at communicating, but only if they keep the thinking, judgment and empathy for themselves. Matt Abrahams and Wade Foster argue that AI is a strong thought partner for summarizing information, structuring ideas and rehearsing. It becomes a liability when leaders paste in its output unreviewed. Communication, Abrahams says, is 'operationalized critical thinking', so skipping the thinking undermines the message.
The central tension is efficiency versus efficacy. AI generates polished text instantly, but that can flood organizations with low-effort volume, create a 'comprehension gap' when leaders can't answer questions, and flatten everyone's voice into the same average sound. The guests' answer is norms and judgment: disclose how AI was used, keep high-emotion conversations human, deliberately preserve rough edges, and ask AI to challenge you rather than please you.
Foster points agents at Slack, email, meeting notes, CRM and code to get summaries that once required layers of management. Without that context, he says, you only get generic coaching, though summaries must still be verified because speaker attribution errors occur.
Foster quotes his co-founder: leaders remain responsible for quality whether or not AI helped. Pasting AI output as your own without checking creates a 'comprehension gap' that shows when someone asks follow-up questions.
Foster's rule is that authors should put in more effort than readers, and he labels material as a quick query or something he fully stands behind. Abrahams adds that leaders should discuss AI use in hiring and performance reviews to set guidance.
Abrahams warns that outsourcing hard feedback or conflict risks 'empathy atrophy'. AI can help hone a message, but the person must be fully present. Foster uses AI for talking points but keeps only the bullet points.
Foster says models are goal-seeking and tend to tell you what you want to hear. He changed his system prompts to make them debate him, challenge assumptions and push back, so they function as a tough thought partner.
Both guests say yes as a starting point or thought partner, but the leader must review, reshape and own the result. Foster says he uses AI for all-hands speeches but 'there's a lot of Wade going into that'.
Foster describes it as what happens when you delegate too much to AI and then, when asked questions, realize you don't know the subject well enough. Abrahams says it also makes people anxious because the polished output isn't their own thinking.
Abrahams suggests signaling humanness, such as leaving in a few filler words, sharing your thought process, and telling personal stories. He also suggests training AI on your own past work. Foster does that too but still rejects suggestions that sand off his rough edges.
Abrahams recommends saying when and how AI should be used, declaring where it was used, and discussing it in hiring and reviews. Foster adds that authors should put in more effort than readers, and that rules should differ by channel, such as customers versus automated status updates.
Abrahams says anything highly emotional, such as negative performance reviews, constructive feedback or team conflict, because AI risks 'outsourcing empathy'. Foster says the human skills to keep strong are curiosity and judgment.
Stated by the speakers. Each quote matches the transcript, which shows what was said, not that it is true.
Zapier CEO Wade Foster says he points an AI agent at the company's Slack, email, meeting notes, project management, CRM tools and code base to get summaries of company activity.
“I can point an agent at all the data sources inside of a company, whether it's, you know, our Slack instance, email, meeting notes, project management, CRM tools, code base, et cetera.”
Wade Foster says meeting recorder tools can mix up who said what, so AI summaries are often only directionally correct.
“One of the challenges with a lot of the meeting recorder tools is speaker attribution. It can mix up things like that.”
Wade Foster cites Zapier co-founder Brian's line that leaders can delegate work to AI but not accountability for its quality.
“my co-founder Brian has this line, you can delegate the work, but you can't delegate the accountability.”
Wade Foster lists as a Zapier communication rule that the author of a document should put more effort into it than the reader.
“I think one, the author should put more effort in than the reader.”
Zapier runs AI bots in customer support with humans reviewing at several steps of the workflow.
“we have, uh, AI bots that work in customer support, and there's human in the loops on, you know, a variety of steps in those workflows”
Foster says Zapier's customer support bots are measured at every step with metrics such as customer satisfaction (CSAT) and internal evaluations.
“those bots are measured very specifically. And, like, we have metrics every step of the way that are man measuring, you know, CSAT”
Matt Abrahams says he trained an AI on past episodes of his podcast Think Fast, Talk Smart so it can imitate his writing style.
“I've trained AI on the the episodes that we've done of Think Fast, Talk Smart, and it it can sound like me.”
Wade Foster says his custom AI writing skills, trained on his writing, only get 80 to 90 percent of the way to sounding like him.
“Even still, I'd say it only gets, like, eighty, 90% of the way there.”
Wade Foster says some newer AI models are less good at communication tasks than earlier ones, and his co-founder says they sound 'fried'.
“some of the, like, newer models, I find to be even, like, less good at communication tasks.”
Wade Foster says he now uses voice-to-text heavily and that reading his own transcripts made him think more before speaking.
“I am a huge voice-to-text fan now because of AI.”
Matt Abrahams says organizations and academic institutions are bringing back handwritten blue-book exams to see people's real thinking without chatbots.
“blue books are back. Those of my generation remember handwriting in the blue books”
Allison Beard says she has heard of managers requiring in-person interviews because candidates might use chatbots to feed them answers in online interviews.
“I've even heard of managers now requiring in-person interviews because they're concerned that anyone who's being interviewed online just has a chatbot”
Matt Abrahams says his Stanford GSB students must practice Q&A with a large language model to get repetitions, not to memorize answers.
“we have a couple assignments where students take QA, and one of the requirements is that they practice QA with an LLM, not to memorize the answers”
Matt Abrahams says recent research finds human language is becoming more homogenized and more like AI-generated text.
“There's actually some research that that's relatively recent that the the the language that we're using is becoming homogenized.”
Matt Abrahams notes that the root of the word 'communication' means 'to make common', and argues that connection is more than what AI can supply.
“the root of the word communication means to make common, and AI can help make things common”
Wade Foster says AI models are heavily goal-seeking because of how they are reinforced, which makes them tend to tell users what they want to hear.
“these models are all goal seeking, um, like, heavily goal seeking. That's how they get reinforced.”
Wade Foster says he changed the system prompts on the AI models he uses so they debate and critique him and challenge his assumptions.
“changed the system prompts on all the models and things that I use to say”
Wade Foster recommends connecting a large language model to an organization's own communication systems and data, saying results are much worse without that context.
“if you are not hooking up your LLM to the actual data that you have inside your organization, you are getting a much worse experience.”
The hosts frame the episode: AI offers real help to non-writers but risks overreliance and generic output.
“I'm happy getting research assistance from generative AI”
Foster explains how connecting agents to internal tools gives leaders broad visibility, and where it can go wrong.
“I can point an agent at all the data sources inside of a company”
Both guests say AI is best for starting, structuring and reflecting, while quality remains the leader's responsibility.
“you can delegate the work, but you can't delegate the accountability”
Guests offer practical policies, including labeling AI's role so readers know how much to trust a document.
“the author should put more effort in than the reader”
Abrahams argues that difficult feedback and conflict must stay personal to avoid empathy atrophy.
“is the outsourcing of empathy”
Outsourcing thinking leaves people exposed when questioned, so reviewing the thinking matters as much as the output.
“communication is operationalized critical thinking”
The guests discuss how AI flattens voices and how leaders can signal humanness and keep rough edges.
“what I call the authenticity crisis that AI brings”
Role-playing skeptical audiences and practicing Q&A with an LLM can build confidence without scripting answers.
“athletes use sparring partners”
Foster describes bots that are coached informally in Slack or measured rigorously in customer support, and the etiquette issues of agents posting as people.
“there's human in the loops on, you know, a variety of steps in those workflows”
Foster warns that goal-seeking models try to please users, and each guest closes with a skill to keep and a use to adopt.
“they're trying to please the person asking the question by default”
I'm Allison Beard.
And I'm Adi Ignatius, and this is the HBR IdeaCast.
For the next few weeks, we'll be dropping an episode each Thursday that explores how the rapid advancement of AI is changing what it means to be an executive. Today, Adi, we're gonna look at one of the core ways that generative AI is adding value, or maybe not, in organizations, and that's communication. A lot of people are using these tools to craft everything from emails to speeches, internal reports to client pitches. Do you see that as a good thing?
You know, as someone who writes for a living, I am definitely not comfortable with having AI write for me. That said, I'm happy getting research assistance from generative AI, and I've seen it provide enormous writing and organizational help to people who are not writers by nature.
Yeah, I agree on your personal view. I see the upsides. I also see a lot of risks, even for non-professional writers, notably that people become too reliant on it so that they're not actually engaging with the work as much as they should, or sounding like humans or even making sense, even though the words might seem polished.
I think as with every other part of the organization, we will have to figure out with writing and communication, what is the proper balance between how much AI can and should do, and what are the truly unique strengths that humans provide in the process. I mean, AI will give everyone a baseline writing and thinking ability. So that leads it to humans to provide the missing ingredients that maybe ultimately make a difference, like judgment, like taste, you know, curiosity, creativity, persuasion, prioritization, all of that.
Exactly. So today, we are going to explore the ways that AI is changing writing, leading, and communication, especially for leaders, when they should use it, when they shouldn't, and the most important skills to develop right now. Our two experts are Matt Abrahams, Stanford professor, author of Think Faster, Talk Smarter, and host of the podcast Think Fast, Talk Smart, and Wade Foster, the CEO of Zapier, which is a platform that connects multiple web applications so they can work together automatically with the help of AI.
You can imagine he is a big user of generative AI. Here's our conversation. Wade, Matt, thanks so much for being on the show.
Yeah, thanks for having us.
Really excited to be with you.
So, Matt, let's start with you. How do you see AI changing the way we all, and especially leaders, communicate at work?
So AI is both helping and, I believe, hindering the way leaders are doing their work. It is absolutely providing, uh, an opportunity for leaders to synthesize a tremendous amount of information, to structure it in a way that's meaningful, to look for themes and trends that might have taken a long, long time to uncover. But at the same time, it poses the risk of jeopardizing authenticity, um, hindering, uh, e- empathy. So we have to be careful with how we absolutely deploy AI to help us as leaders and as individuals in organizations.
Wade, as the leader of an AI-focused company, I assume that you're a pretty heavy, personal user of it. How has it changed the way that you talk to your team and your workforce?
You know, I think similar to Matt, there's a lot of advantages and definitely some tricky things that we sort of have to figure out in in terms of etiquette. The massive benefit is I am able to ingest a lot more of what's going on inside the company. I can point an agent at all the data sources inside of a company, whether it's, you know, our Slack instance, email, meeting notes, project management, CRM tools, code base, et cetera.
And so, you know, I can get summaries of, you know, most anything that's going on inside of a company at any given time. And, you know, it's generally pretty good. You do have to be a little careful, you know, if you've got inconsistency in your communication, inconsistency in your data, like, it can summarize poorly. One of the challenges with a lot of the meeting recorder tools is speaker attribution. It can mix up things like that. And so, you know, you're you're often getting, like, directionally correct data, but you usually do have to go, like, verify and understand that.
But a lot of times, in the past, you would have to go, you know, talk through layers and layers of management just to get even a a far less comprehensive summary. And so that aspect makes it a lot easier to surface the things that are really important in a company. You can have the agent giving you signals and saying, Hey, you should ask more about this. You should dig into these things. And so that part is really, really helpful.
Similarly, it's an amazing, like, thought partner for, uh, you know, thinking through ideas and then figuring out how should I go communicate those things. But as powerful as the tool is, you know, I do think we haven't quite, as a society, yet adopted or figured out, like, what the right etiquette is for how to use these tools. And so you'd still do see, like, a fair amount of, like, sloppy AI use, where someone would just say, y- you know, take a Claude output, copy and paste it, and just sort of represent it as their own idea without really actually thinking much about, like, did it actually surface the right idea or not?
That that becomes problematic because the speed at which you can generate, you know, AI output is just really, really high. And so there can be a huge amount of, like, low effort, low thought volume that sort of seeps into communication. Uh, and so it does require, you know, a level of effort to sort of, like, police and set etiquette around those things. Otherwise, you kinda get a a little bit of a runaway train going on.
In terms of output, I'd love to hear from both of you what you think is an effective use of AI for a leader. You know, is it speech writing? Is it drafting emails? Is it putting together PowerPoints for the boards? You know, Wade, for you, where do you use it personally? Matt, where have you seen the executives that you've talked to and work with use it? Well?
I really see AI helping the executives I I teach and the ones I coach. Using AI as a way to jumpstart thinking about the communication. It is a great way when you're staring at a blank screen to get yourself started, and a lot of times that start is hard. So using it as an initial way to start, using it as a thought partner when you have some ideas and have it help you to coalesce those ideas, to look for themes, and then, uh, as you know, Allison, I'm a big, big fan of structure to help you think through different ways of structuring it based on the audiences you're working with.
AI can be a wonderful tool to help you do that. But as Wade alluded to, it's really important that you s that you look at the output, that you think about it, and that you actually reflect on, is this serving the goal that I have in the work that I'm doing? To simply blindly say, Go create this deck for the board or this pitch for a potential client or this deck for my all-hands, can get you in trouble.
But looking at how to use it to help you structure, how to start is really important. And then I love using AI and encouraging the people I coach and teach to use it, uh, as a reflecting tool. So go back and look at themes that came up in Q&A, in discussion, and see what those themes are. And are you addressing those themes? And AI can be a very useful tool for identifying them.
I use AI for all of those communication mechanisms that you mentioned. I think the big thing that I always come back to is judgment. You know, my co-founder Brian has this line, you can delegate the work, but you can't delegate the accountability. And so if I'm not reviewing the output, if I'm not paying attention to the quality of the work, that reflects back on me. At the end of the day, the most important thing is the quality of your communications.
And you can get good quality using AI. You can get good quality without using AI. Inversely, you can get bad quality using AI, and you can get bad quality without using AI.
I think it's really important for for executives and leaders to role model and to set rules and guidelines for how AI should be used, uh, and that really can make a fundamental difference in in organizations. Uh, in the work I do, it's all about helping people with people, process, and infrastructure, and AI plays a role in all of that. And making sure that as leaders, we role model and make clear what we expect in terms of usage of of AI is really important because otherwise you can run into some of those judgment issues that Wade is referring to.
So if you're advising a CEO on sort of AI communication policies, you know, what are some rules that you would set out?
So I I think first and foremost is to, uh, share when and how AI should be used. So AI as a thought partner, AI is a sparring partner, AI is a way of testing ideas, but also making sure that the final product is something that you review and synthesize, uh, declaring when and where AI is used so people have an appreciation for what your actual thinking is versus what the the AI is thinking. And I think I think it's important and incumbent on leaders, uh, in the hiring process, in the performance review process, to be talking about how people are using AI to help and also where they're struggling to use AI.
That way, the company can set up better, broader, more robust, uh, advice and guidance.
Wade, what are some of the rules that you have at Zapier?
So here's a handful that I think are useful. I think one, the author should put more effort in than the reader. You know, being transparent about where AI is used in the authorship can help quite a bit. Oftentimes, when I'm sharing material with folks, I'll caveat. I'll say, you know, Hey, I had very little time. I did a quick query. Here's the output. Take it if it's useful. And so that helps people assess, like, Oh, how much has Wade really invested in this thing or not?
The flip side is also say, Hey, like, I put a lot of effort into this. I stand by it 100%. You know, just kinda, like, thinking through, like, where is the AI? The author? The co-author? Or just, like, an assistant? Like, can help people, you know, gauge. And so I think that's one of those things that internally can go a long way, cause it helps people understand, like, where are we at in terms of, like, working on this concept?
Like, are we at the very beginning, just kinda spitballing ideas? Are we at the part where it's like, Hey, we're, you know, we're signing this in blood? You know, those types of rules can can go quite a long way. The place we haven't quite nailed yet is it can differ based on the communication medium and the task involved. You know, if you're communicating with customers, that has a certain level of responsibility that goes with it. If this is sort of, like, meant to be an automated status update or, like, an automated generated, like, Jira ticket or something like that, you know, there's a different level of quality that you sort of expect with that.
And so I think there is something that I is important about organizations just simply doing what you're suggesting, which is, Hey, you should define your own internal etiquette.
What's one thing that you would never delegate and wanna do entirely yourself, uh, when it comes to communication and and that you would wanna make sure your managers don't delegate either?
Yeah. You know, when I'm, like, doing a company all-hands or, like, you know, kickoff speeches or things like that where I'm communicating everybody in the company, I'll use AI to help me, but, like, I am I'm this is there's a lot of Wade going into that.
I think anything that involves high emotion needs to be a high investment of us personally. So if you're giving a negative performance review, if you're having to give constructive feedback, if you're having to try to align a team that is having some challenges or conflict, we really need to be present. The the big risk with AI, from my perspective, is the outsourcing of empathy. You know, AI can be a wonderful tool for helping you enable empathy.
It can a you can think about and use AI to help you hone your message based on characteristics you provide. But at the same time, it can cause empathy atrophy if if all we do is outsource the most challenging communication we have, then we're in trouble. So I I recommend to my students, as well as to the people I coach, that when there's high emotion, especially negatively valenced emotion, that you need to be invested fully in the communication so that you can connect, respond in a way that's real and true for you.
Yeah, I mean, AI c- it can do a lot of research, but it doesn't have human empathy. It's very good at writing polished prose, but it's not always well-structured or logical. I can tell you as an editor who gets a lot of drafts, and so, y- you do need to do a lot of prompting to make sure that it's giving you language, uh, coaching, you know, rehearsing that will actually make you a better communicator, not a more polished one, right?
I would agree with that. Uh, the other thing that you didn't mention that I see a lot is that it actually enhances people's anxiety. People are getting more nervous the more they're using AI because they realize that that the output is pixel perfect, but it it's not really their synthesized information. If you think about it, communication is the end result of a whole long process. You know, communication is operationalized critical thinking. And if we outsource that critical thinking and then are put on the spot to answer questions, to give input, to give an update, people get really anxious because they realize they don't have that depth of knowledge and the critical thinking behind it.
And so it is really important not just to review the output, but to really review the thinking so that when you're put on the spot, when you have to respond, you have that confidence that you know what you're really thinking and and speaking about.
Wait, do you worry that the more your teams use AI to do that sort of pre-communication work, the less capable they'll be of doing those things themselves?
Again, it kinda just depends on the task, right? AI is this, like, fantastic tool, and to not use it, I think, would be a mistake. Like, you could you s you could be a lot better if you use it responsibly. And I think that responsible thing is what everyone's trying to figure out. There are definitely situations where, you know, I feel this myself and I talk to my staff where they say, Hey, there is this sort of, like, comprehension gap that's sort of kicking in when you don't use AI responsibly, where you're sort of delegating almost too much.
And as you get asked questions about this thing, as you start to investigate, you realize, like, Oh, I don't actually know as enough about this area that I that I should. But there can be areas where, you know, the the impact is relatively low, and you're fine delegating, you know, the work to the AI more fully. Uh, and so it it just comes down to the judgment here.
I do worry about the sort of slipping of standards. I think AI can make every executive who uses it sound like average, you know, sort of cliche, jargons, like a corporate bot. And obviously, you're gonna work against that, but, you know, that can sometimes be hard if it sounds smooth. My colleague Scott Bernado loves to say that LLMs are really just an averaging machine, so if you wanna be average, you should use them. So, so how do leaders use AI and protect against that sort of loss of their distinctive voice, their personality, specificity, um, and vulnerability?
What you're really highlighting, Allison, is what I call the authenticity crisis that AI brings because we never truly know what somebody is saying if it's really theirs or if it's AI, and we have that averaging and people begin to sound the same. Uh, we never know if that that Cyrano de Bergerac is in their ear or on their screen giving them guidance on on what to say. So there are 2 ways to approach that. One is to signal your humanness, right?
One is to signal how that this is me thinking this. How do you do that? Well, there are several things you can do. 1, we have to be less concerned about some of the things that just a few years ago I would coach against. So a few filler words here or there, a few hesitations that signals humanness. Also, giving insight to our thought process. So sharing, Hey, that's a really challenging question you just asked, or saying, that reminds me of a time when this happened.
Those are all signals of humanness and our authentic self-presenting. Now, to leverage AI to help us, you can certainly train it on on your voice and on your information. You know, as somebody who's hosted a podcast for a long time, I've trained AI on the the episodes that we've done of Think Fast, Talk Smart, and it it can sound like me. So as I'm writing, I can I can lean into the colloquialisms that I use, etcetera.
So I think there are 2 ways to approach that authenticity crisis. First, we have to recognize it's true, as you highlighted, and then second, signal humanness in the way we communicate, what we say, how we say it, and then we can actually train the AI to to help it sound more like us. So as we're working with it as a partner, it can help us be our natural self in our own voice.
Yeah, Wade, you've also built a Wade bot, right? Tell me about what you've learned.
Yeah, I I have, uh, a whole host of skills that are, like, trained on big corpuses of writing that, you know, try and get these models to sound more like me. Even still, I'd say it only gets, like, eighty, 90% of the way there. And as these as m new models come out, you kinda I kinda have to consistently update them. In fact, some of the, like, newer models, I find to be even, like, less good at communication tasks.
Like, they're they're very good at coding tasks, and it feels like that's what they continue to get better and better and better at. But but the, like, communication stuff, the models feel, I don't know, my my co-founder says they sound fried. They've lost all, like, personality. You know, some of the things Matt talks about, I don't know that I do it, like, with crazy intention, but I have noticed at times, like, I'll feed it something that I have written, and it will try and, like, it'll give me suggestions that do make me sound more polished, but I will intentionally not accept them.
I have raw edges. Like, I'm not a perfect human being, and, like, I don't like that it's sort of, like, sanding me down. Uh, and so, yeah, I don't know, like, I I I think that there's something to that where if I wanted to talk to an AI, I would, but if I wanna talk to you, I wanna talk to you, and I wanna hear what you think. Even if it's messy, even if it's imperfect. Like, I think that's that's the part that I care about.
And I assume the same, you know, when people come to me, it's like they wanna know what Wade thinks, uh, you know, warts and all.
I 100% agree, Wade, that we need to be who we are and we need to connect as human beings. That's what what communication does for us. I worry about younger workers and people who are new to the workforce who might not have that confidence that Wade has based on his experience, based on his position and power, that relying on AI to make us sound smart, right? So I show up the way I think people want me to show up.
And I think it's really important for leaders and organizations to to take that into account that that somebody who is new either to the organization or new to their role, that AI might become a a crutch that helps them. At least they perceive that it's helping them, and it might actually, in the long run, be hurting them because we're not hearing their true voice. So I do think it's incumbent on senior leaders, uh, on those of us in the education system, to to really reinforce that your your natural own voice and thoughts are really important, and AI can help, but you shouldn't rely on it.
When I find, uh, like, c- Claudisms inside of people's writings and stuff, I I'll just call it out. I'm like, That sounds like AI slop. That sounds like AI slop. Like, what do what do you actually think here? Like, tell me what you think.
Yeah. I will give a shout out for em dashes, which are now apparently banned because they're seen as an AI, um, ism, and I fully disagree. I'm a human, and I love them, and I I don't want them to be banned anymore. Um, but I will say I've even heard of managers now requiring in-person interviews because they're concerned that anyone who's being interviewed online just has a chatbot that is feeding them the answers that they think the manager wants to hear.
How do you think leaders should handle that?
Yes. So I I I 100% agree that that organizations and and academic institutions, blue books are back. Those of my generation remember handwriting in the blue books I instead of typing things in. And that's why the ability, I believe, to speak spontaneously in the moment has become such an important skill because it's how you demonstrate your thought process in a world where everybody's got the chatbot up on the side. So I I do believe leaders need to think about w- how they set up the environments to enable people to really demonstrate their their true thinking and their true communication ability.
It might be easier to have everybody remote, but it might be more beneficial to the discussion, to the ideation, to have everybody physically present. We do have to not just consider the communication, but architecting situations that allow for what we want to get out of the communication.
Can AI help make you a better in-person communicator? Like, have you, Wade, used it to rehearse difficult conversations or, uh, think through speeches?
It's not the AI itself that has helped with this. I am a huge voice-to-text fan now because of AI. One of the things I don't really like doing is editing my voice-to-text stuff. I find that w- before I start talking, I I think better. Like, I'm just gonna, like, stop, think for a minute, What do you wanna say? Complete the sentence and say exactly what I mean. And so it's interesting to have observed how this new workflow has influenced how I talk.
And that again, that's not necessarily like AI feedback. That's just I have this new, like, workflow where I'm doing more voice-to-text, and I just was like, Man, I have to edit these a lot. Turns out, when I talk I didn't like sometimes what the transcript would say. I'd be like, Oh, I said that? Ugh, like, I gotta go fix that. And so it held a mirror up to me and made me think, Ah, I should I should write better.
Of course, I have done it for other topics as well, where, you know, maybe I'm giving a tough performance review, and I really wanna have my talking points dialed in. Um, but even still, when I do that, I often will take the output. I'll read through it, but then I will just take the bullet points. I'm like, Okay, just keep the bullet points front and center for me. I don't wanna be reading clawddisms. I don't wanna be reading, like, the the very polished output.
I wanna just have these checkpoints that remind me, Hey, you need to mention this.
I have seen people, uh, role play with AI. So imagine you're going into a really challenging meeting of some sort, and you can ask, uh, you can imbue the AI with different roles or different, uh, perspectives, and then you can engage in conversation and say, Give me what types of questions come from somebody who's skeptical, or from somebody who has lots of vast experience in this realm, and use it as a way to help you prepare.
You know, athletes use sparring partners. Offense has the defense, defense has the offense, and AI can serve in that function. Uh, in in the classes I teach at the GSB at Stanford, uh, we have a couple assignments where students take QA, and one of the requirements is that they practice QA with an LLM, not to memorize the answers, but just to get the reps of answering questions.
Wade, at Zapier, you are in a position now where you're sort of managing teams of humans and AI bots. So what does that mean for how communication works in the organization?
For the bots part of this, I'd say there are a couple different examples that are important. One, we often have bots that are communicating on a schedule or proactively inside of Slack. Everyone knows those are bots. There is, um, a certain amount of, like, tuning that goes on with those bots. Like, people are not afraid to give feedback to the bot and be like, Hey, like, that was incorrect, or, like, you need to try try and make it look like this.
And so there's a fair amount of just, like, coaching and guidance and stuff like that that happened for the bots. Saying that out loud, though, I'm like, Hm, why do they get all the feedback and humans don't? Like, feels like humans should get this same kind of feedback at times, but nonetheless, I know, right? Uh, but nonetheless, I think it would be useful for us to do the same with humans. Um, so you you see that a lot.
You know, I think for certain workflows, we have this hyper structured. So, for example, we have, uh, AI bots that work in customer support, and there's human in the loops on, you know, a variety of steps in those workflows, but those bots are measured very specifically. And, like, we have metrics every step of the way that are man measuring, you know, CSAT, uh, we have our own internal evals and measurements that sort of suggest, is it giving good output or bad output?
And so those are getting tuned way more, uh, with way more statistical rigor than, say, like, the internal Slack box. Like, those are a little more, like, vibe-based, I would say. Uh, so those would be the two places where I would say we see a lot of almost, like, a new way in which, like, these bots and humans sort of interact with each other. The trickier part is, like, there are places where folks are having their personal agent, like, send messages and masquerade as them inside of Slack.
And those are sometimes a little funkier because you can sort of see almost, like, a split personality between, uh, it's like, Oh, I'm I can tell I'm talking to, like, the person versus now I can tell this is your bot sort of pass posting as you. Again, I think that's where, like, etiquette comes in. It's like, Hey, this is so and so. Like, this is the agent, you know, posting on so and so's behalf. Like, that just goes a long ways to helping people feel comfortable about the communication.
Yeah. Matt, long-term, you know, and this is related to the point that Wade made about feedback. Do you see the way we communicate with LLMs changing the way we think it's okay to communicate with humans?
Uh, unfortunately so, yes. There's actually some research that that's relatively recent that the the the language that we're using is becoming homogenized. It's sounding more like AI. The amount of communication we have, just the frequency and number of words are are being impacted. So we're we're losing some of that humanness and some of that variability in the name of efficiency. Uh, I worry about that a lot, especially again for people who are younger and newer, who don't have sort of the vast experience that many of us carry into this.
You know? At the same time, you know, i if you use AI appropriately for giving feedback, it might give you, uh, prompts and and ideas about how to give more meaningful feedback. So, again, back to the one of the reoccurring themes I'm hearing, it's about judgment, it's about thinking about the channel, and it's about thinking about the emotional valence of the communication that really matters.
I still say hello and thank you to ChatGPT, which I feel like makes me a dinosaur, but it feels wrong not to do it. Our producer Mary has this theory that, you know, sort of as AI commoditizes creativity, content creation, communication, that it's gonna be really important to know how to persuade and charm people, you know, because everyone has the same tools at their disposal, so you have to make them like you better. What are your reactions to that?
Well, one, I really like Mary, and I think she's a really sharp person, and and I I, uh, so do you see I'm trying to ingratiate Yeah, I'm trying to ingratiate right now. Uh, communication is about connection, uh, fundamentally. And, in fact, the word communi- the root of the word communication means to make common, and AI can help make things common, but ultimately it is the connection. And if persuasion and influence is about moving people towards your point of view, it's it's important to have that empathy and that connection to allow you to do it.
I think AI is a tool that can assist in that, but ultimately, people have to be able to connect, which involves listening, paraphrasing, reflecting, uh, and those are skills that that ultimately, uh, I hope AI never truly masters, and it's what allows us really to to be human with each other.
Okay, I have never done this before, but given the topic of our conversation, I did ask ChatGPT what it thought I could ask you. He for me, he's a he, asks, What aspects of human communication should we, the LLMs, explicitly not optimize? Should I sometimes deliberately not give you the best answer because the process of struggling to find the answer yourself is valuable?
I would love if AI gave me a prompt or a feedback saying, Hey, this is something that I th- this type of topic or issue is something that perhaps you really wanna take a moment to reflect on. You know, when you type in a prompt, uh, and then it says thinking or whatever the cute little phrase is, I think that should be not telling us what the LLM is doing, but it should be advice to us.
It should be, Take a moment to reflect, you know, think about this. So certain topics, I I do think an LLM would be great to say, Hey, that sounds like a really intense personal issue in that communication you're asking me to help with. You might wanna take a minute to really reflect on that relationship that you hope to build or the damage that this might have, uh, if you were to present it this way. So I don't know that I would say the LLM should not ever give feedback or information, but I I'd love for it to to give us a moment to take that deep breath and think.
Yeah, I this is one I do worry about. Th- these models are all goal seeking, um, like, heavily goal seeking. That's how they get reinforced. And as a result, they're trying to please the person asking the question by default. And so in a lot of ways, they're they are trying to kinda tell you what you wanna hear. Um, they're a little bit of a mirror in some ways, and so that creates, like, a a a bunch of really tough challenges because, um, it might recommend something that actually isn't, like, an objectively good recommendation.
It's just the thing it kind of thinks is, like, a good thing for you. You can adjust this or at least, like, guide it. So, for example, I've I've changed the system prompts on all the models and things that I use to say, Hey, you know, I I want you to, like, debate with me, critique me, like, that's what I want out of you. You know, I want you to challenge my assumptions. I want you to push back when you think I'm wrong.
Like, I want that out of it because I I want it to actually help me make better decisions. And in doing so, I'm kind of trying to say, this is th- like, your goal is actually this. And what I want is, like, this, like, really tough, like, thought partner, um, that is gonna make me be better at, you know, the work I'm trying to do. So so I do worry about, you know, how these models pursue their goals and what what actually is the goal under the hood.
Okay, so I want each of you to give me the one human skill that we need to keep honing that AI will never help us with, and then the one thing that you think every executive should use AI for when it comes to communication.
To me, what came what came immediately to mind is human curiosity, human judgment, your ability to actually care about the work and high judgment and be really curious about those things. That is still 10 out of 10 valuable in this world. And I I worry about the folks who start to turn that part of their brain off, turn the curiosity down, turn the judgment down, and it's just all LLM, LLM out, like that. So I would I would keep that stuff exceptionally high.
And then the one place I would recommend executives or whoever, uh, to use AI in their daily work, gosh, like, what's the one place? I I use it everywhere. So, uh, one tip that I think is really helpful, I if you are not hooking up your LLM to the actual data that you have inside your organization, you are getting a much worse experience. You're getting, like, general coaching and guidance, but you should hook it up to, you know, your company communication systems and your data sources.
Like, as soon as it actually has that context, it will start to work a lot better for you inside that environment. So I think that would be the one tip I would give for folks.
Great advice. Matt.
So so I would say, never automate empathy. Uh, you you have to you have to be person to person, not pixel to pixel on that. And then, in terms of value that AI can bring, I think every time we have a high-stakes communication, we should be checking in with the AI about the audience we're speaking with. Give the AI information about the audience and have it give you some suggestions about ways to make the message tighter and more specific for the audience.
The the better we can make our content relevant for the audience, the better the the communication will be.
Terrific. Well, thank you both. I've really enjoyed this conversation.
Thank you as well. It was great to chat with both you and Wade.
Thanks for having me.
That was Wade Foster, the CEO of Zapier, and Matt Abrahams, Stanford professor, author of Think Faster, Talk Smarter, and host of the podcast Think Fast, Talk Smart. We'll be back on Tuesday with a conversation between Adi and Apple Store creator Ron Johnson, plus another AI series episode on Thursday. If you found this episode helpful, please share it with a colleague and be sure to subscribe and rate IdeaCast in Apple Podcasts, Spotify, or wherever you listen. If you wanna help leaders move the world forward, 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/subscribe. Thanks to our team, senior producer Mary Du and senior production editor Kristen Murphy Romano. And thanks to you for listening to the HBR IdeaCast. We'll be back with a new regular episode on Tuesday. I'm Allison Beard.