Women at Work - Chats, Bots, and Prompts: Make GenAI Work for You

Episode Date: November 11, 2024

There’s something about hearing how other women are making the most of LLMs that can turn even the most GenAI-avoidant among us GenAI-curious. At least that’s what happened to the Amys when they heard from several power users who’ve broadened their thinking, deepened their agency at work, and saved themselves time and stress. Maybe in listening to them you’ll be inspired too.

Harvard Business School Executive Education develops leaders who make a difference in the world. In their programs, experience the power of fresh perspectives and connect with a world of new ideas. Learn more at hbs.me slash learn. That's hbs.me slash learn. Hi everyone. Like many of you, Amy B and I are taking in last week's election news and thinking through what it means for women and our workplaces. This show's mission is to help each other move forward together, including when things are very uncertain. Today's episode offers support in a completely different area of uncertainty.

Generative AI. We hope you find it helpful. You're listening to Women at Work from Harvard Business Review. I'm Amy Gallo. And I'm Amy Bernstein. Some of us have gone all in on generative AI. Like these audience members of ours. So I created this custom GPT for my job.

It was a game changer for me. Before, if I had a problem, it would take me one day to solve. Right now with these tools, I will do it, I don't know, in three hours. It drastically increases efficiency and proficiency. I am able to be a better leader, produce more work. I have definitely come to rely on it. I'm just so much more satisfied in my work because the opportunity to step back and look at things and translating it into something that's actionable for my teams, that's a huge value-add. Computers now speak our language so we

can use our own English language skills to tease what we want out of these tools. So you hire a fake team of researchers and each one of your fake researchers comes up with a research plan and then what's better is if you create a character who's a behavioral economist on the team, and the behavioral economist will come in and look at all the results of this research and tell you what's unrealistic. It's just unbelievably different and more impactful the stuff you can do. Aim, have you had these revelations in the AI yet? No, not even close.

I mean, I've asked it to do things that I was pretty sure it would fail at, and it did not disappoint me. So that just affirmed for me that I didn't really need to pay that much attention. Wait, what do you mean you asked it things you knew? Oh, maybe three months after Chachi BT debuted, I was writing a condolence letter and I don't know about you but I find those really hard. Hard to get started. Hard to get, actually, just hard to do the middle part and hard to end. But I thought, well, okay, do it. Write me my condolence letter and I came up with

several prompts and I was tweaking the prompts and you know it I came up with several prompts, and I was tweaking the prompts. And you know, it never came up with something that I could put down on a card and send. I'm kind of relieved at that, though, to be honest. That relief, I think, is exactly where I am with this, which is that I recognize it's powerful. I know it could be transformational for me, certainly for my work, definitely for the world. And yet I don't want to accept it. And so when it fails, I'm so happy. I still have a job.

Yay! You see? You see? Well, except these are early days and this technology is going to learn so quickly. I agree. I know it's going to change the way I relate to my work. How so? Say more about that.

Well, there are just aspects of my work that I don't like that this can help with. And we'll get into in a little bit the power users we've heard from because this is who I'm really basing this on. It's the women who wrote into us who told us how they're using it and whether we like it or not, whether we want it to fail or not, it's happening and so we have to be early adopters although we're late on that. We have to be middle adopters. We have to adopt. Let's just put it that way. We have to adopt. In listening to women like us, like you describe how gen AI has saved them time, broadened their thinking, deepened their agency at work, Amy B and I have gone from being skeptical dabblers to being skeptical enthusiasts.

Maybe listening will have the same effect on you. If you haven't yet tried your hand at Gen.ai, we hope these power users inspire you to finally give it a try. And if you're already using the technology, that you come away with some new ideas. One of the things I was really struck by was the mental and emotional relief that GEN.AI gave them. Blew me away. Yeah. Because of course I expect the opposite because that's been my experience.

It makes me anxious. I don't know how to use it. It's going to take my job. And for them to actually be able to Begin to use it in their day-to-day work and almost all of them use it on a daily basis but to relieve mental burdens such as burnout or task paralysis It's just been a huge source of of relief. Can we hear from yeah, let's listen to it let's listen to what Andrea has to say.

I would consider myself a middle manager. So at the time I discovered chat GPT and generative AI in general, I, like many people, was coming off several very difficult years. Difficult personally, difficult from a work perspective, and also very difficult as a leader. Going through the pandemic, we really wanted to be authentic

and present for what was going on in people's lives as individuals. And that's a lot. It was draining. And I definitely had reached the stage of burnout. That's a lot. It was draining and I Definitely had reached the stage of burnout and That's where the generative AI really made a huge difference in that was How much energy do I put into things?

Because there's so much in your day that just comes through right that? That you have to react to, react to, react to. And you definitely need to put some thought into it, but you're not writing the US Constitution. And so rule of thumb, if I'm likely to overthink it, let me think of a prompt, let me think of a way to respond to this that doesn't require as much mental effort or kind of right sizes the level of mental effort to the task at hand. Because the things that like I said shouldn't require so much of our energy often do, the power of being able to just jot down a few thoughts, make an executive summary of this. Boop! Here's three bullet points. Rewrite this for a group of cross-functional stakeholders. Boop! There it is. Make this sound more professional. Make it sound more casual.

Make it longer. Make it shorter. Expand on this point. Em this, define a call to action. Boop! It feels like a miracle. So when I think of the many, many moments of misery I've experienced in my career, I cannot tell you how high a proportion of them had to do with not being able to right size the effort and attendant anxiety to the task at hand. Yes. Well, and to not get stuck in that moment of how important is this thing, and if I know it's not that important, why am I spending so much time thinking about it? And what Andrea is describing is almost a way to short-circuit your perfectionism.

Oh, absolutely. You know, I'd love to hear from another listener who uses it in a similar way but in a different context to focus herself in a way that is more appropriate and will get her to her goal. Jessica McBride I'm Jessica McBride, the founder of Tech Savvy Assistant. I created Tech Savvy Assistant after being laid off from my EA tech job based out of New York.

I used ChatGBT because it was the original one and I realized quickly how functional and helpful it was for admin roles, especially, because there's too many nuances that go into what we're trying to solve that cannot really be encapsulated by a Google search. And so by using ChatGVT, I was able to take these severely nuanced problems and parcel them down into something

that was a much more accessible way for me to take a problem on. A problem that we come across a lot of times in the admin field, is somebody comes to us and they're like, ''Hey, I'd like for you to do an event.'' They give us this big idea, synergy. I would just be like a deer in the headlights. But you can workshop this idea of like, ''Okay,

we need to do a three-day conference based around the idea of synergy.'' Help me come up with a step-by-step guide on how to produce this event. And what I love about it is that, yes, I'm perfectly capable of creating this event on my own, but what am I going to miss? What steps am I not noticing? What else would I like to take into consideration for this? And it's going to give such a well-rounded idea of what you want to create. It's just giving you the roadmap. A lot of people in my industry lashed on this idea

of AI isn't going to replace you and EA using AI well. And I don't like that. I think that that is anti-fellow worker. I believe that we should be using AI to empower our careers. I think it's no different than, you know, getting introduced to the internet. This is the next level of like, I just need to learn how to embrace and use it and utilize it to the best of my ability. And I can really empower my career and become the next generation of executive assistant.

There are so many things that we can do to like increase the culture and be strategic business partners and think and expand that I did by just talking to other leaders in our organization and solving problems. But you can use AI to help you solve these problems. What I heard in that that really excited me was this idea of breaking down the daunting challenge into completable tasks, right? I found myself wondering if LLMs are freeing admins to do higher level work.

How are they doing that? You know, and Jessica has really tackled that. It's incredibly impressive. She has figured out how to use the technology to do a better job of identifying her executives' objectives. And she's actually codifying. Codifying and making it available. All of this and making it available to other EAs, which is just so smart.

I have the guide in my hand, Aime, and let me just read you some of the others. Identifying resources and obstacles. I cannot tell you how many hour-long meetings I've sat through, participated in, where we were trying to identify resources and obstacles. And she has a bot that does that. Developing an action plan. When I read this, I thought, well, this is everyone who has anything to do with formulating strategy and executing on strategy.

And she's, again, making that available to anyone. That's one of the things that I think is really encouraging about these power users is so many of them are not just creating these tools for themselves, but also thinking about how it can benefit their clients, their friends, their community, their network. You can visit techsaviasistent.com and on there you'll find things like my newsletter that goes out weekly that often includes videos and clips of how I'm utilizing ChatGPT. I also have a community that you can join. So it's really focused on like keeping you current and what is happening in technology.

The other thing that I heard in Jessica and Andrea's stories were that they are using it to get unstuck. I want to go back to that for a second because, you know, as you said earlier, the amount of agony we've spent in these indecisive moments, and it's something I actually really worry about with myself is that I tend to be indecisive about certain types of decisions. And I've actually worked with a coach to help me on certain decisions. So like when an opportunity comes to me, whether I say yes or no is so overwhelming to me. And I've actually worked with a coach to come up with the criteria,

when will I say yes, when will I say no? And I go through that criteria, I rank it, but as I'm listening to Jessica talk, I'm thinking, why don't I make that a bot? Why don't you make that a bot? And just, I already have the criteria. I can feed it decisions I made before, so it knows what I've said yes to. I can also feed it all of the sample language

I've developed around saying no, because that's often the hurdle, is that I decide I'm going to say no, but now I have to write the no, and that's really uncomfortable for me. So I can feed it that sample language, and I can say, should I say yes or no to this decision? Why? Using my criteria, tell me why. Can you please write a note saying no if the answer is no, it often is. Yeah, and it really brings home to me how we're really only limited by our own imaginations.

How we could use this technology. And you describe the steps. It's not that complicated. It's not. But if someone said, oh, you should develop a bot that will help you make decisions, I would, two weeks ago, I would have said, yeah, that sounds great. Never. Like, I don't have the time for that. Will the amount of time I'm going to put in actually benefit me? Like, no. But listening to these women describe how they're using it makes it seem so much more doable. Yeah. And it helps externalize that anxiety. If you do it right, you're going to get to the same conclusion, but you want to put yourself through all the stress of getting there. Right.

We have some other users who have been using it as sort of thought partners. And I'd like to hear from them in terms of how are they using it to both get unstuck, but also helping them generate ideas and think creatively. Because I think that's one of the things we've often thought it wouldn't be able to do, is creativity. But let's hear from Pam first.

So for example, I am working on an account-based marketing program for a client, and I'm a marketing leader, so I've done a lot of account-based marketing programs in my day. But it's really easy to sort of get stuck in a rut. So I actually put a prompt and said, I am working on an account-based marketing program for this company. We want to save these types of messages. What are some tactics that I should think about? What are some approaches that I should think about?

And it actually came back with a list of the personas that I might be contacting. And the kind of company that I'm working for is in the HR space. So I'm used to speaking to chief human resources officers and heads of recruiting and heads of talent. But one of the suggestions that it came back with was heads of operations. And I said, oh, that's really interesting. I've not really worked with that persona before.

So I kind of furthered that query and said, okay, tell me about heads of operations. Tell me about that persona. So the suggestion that it came back with is heads of operations are really thinking about costs and cost savings and to consider building a calculator based on some very specific criteria and metrics related to operational costs. It would have never occurred to me to do that. So whether we actually do that, and I could have, and I probably will at a separate time, continue that conversation with the AI to tease out other ideas for reaching out to that buyer persona. What I like about that use case is that because one of the hesitations and one of, I will

now say an excuse that I've been using about not using GenAI as much is that it's not accurate. That's a great case where it doesn't need to be accurate. It's just suggesting something she might explore. And in partnership with her, continuing to explore it. So, you know, she even says whether we actually do that or not, that's yet to be seen, but it's a new idea for me to think about. Right. That's creative. Yeah. And to think about the sources of input I get on a project I'm working on, right? Who am I not reaching out to to get that input? Or I have a target audience when I'm writing

an article. Who am I not considering part of that target audience that might be it? I mean, it used to be we had to spend tens of thousands of dollars, maybe more, and months to get that data about potential readers. But now we can begin to imagine who else might we include. And that's going to change the way we think about ideas, the way we market the products we create. The thoroughness of the consideration. Could you always worry about what you're missing? Yes.

Everything we're talking about here, while we're a little bit starry-eyed, it's also making me uncomfortable. I don't know. Sure. Yeah. Because if these agents can do such a good job, then what do we need us for? And that's that is a huge question. It is. And unanswered. Unanswered. So I'm gonna

say a year and a half ago I was recording a conversation with Kareem Lakhani, who is one of the experts on AI, and particularly a generator of AI over at Harvard Business School. And he was saying to me that, you know, it would take almost nothing for him to create an Amybot, I edit him. He's got a lot of correspondence for me,

he's got all of the marked up drafts. And he said, yeah, it really wouldn't be hard at all. Just recounting this to you is giving me the chilly horrors, right? Yeah, because I know he could do it. Right. Well, but why didn't he, I wonder? Why didn't he? Like, is it out of respect? Or is there something he knows? Oh, he's got way more important things to do. Or is there something he knows? Oh, he's got way more important things to do with it. Well, maybe that's it. Maybe that's it. He's like, I'll let Amy

keep her job, because I'm too busy. Thanks, Kareem. Alright, well let's talk about another way that these women are using AI. And in particular around professional development and career empowerment.

So I am Barbara Ruiz. I'm an SAP consultant and I started using AI because my dad taught me. It was a game changer for me because in my role as SAP consultant, there is a lot of deep analysis, complex problem solving. So I created this custom GPT for my job. I put how I wanted the GPT to present me the information and then also where I wanted the information to be retrieved from and also some PDF that I had with knowledge I have been gathering through the years. I now solve technical issues that before they were like outside of my expertise because the GPT will guide me step by step. And even if it won't give you the solution,

it can give you very good hints of how you can get to the solution. It's faster. It requires less people also because then I don't have to go all the time to developers. And I feel honestly is very empowering because now I don't need other people to help me. I'm just like, I can do it by my own. I just I'm very happy to be able to to improve my efficiency

and to have also a bit of free time because that also have helped me to to have free time. The thing Barbara said about working more independently and not having to call the developers, I found that really both exciting and terrifying. What happens when we don't need other people anymore, Amy? And this is one of the things I always worry about in new technologies that promise efficiency, is that sometimes the most efficient way isn't the best way.

And there's a certain amount of friction that we need to generate new ideas to form relationships, to build trust, to resolve conflicts that I worry we're trying to sort of smooth the edges of the messiness of work in a way that's going to end up dampening our humanness. But I think that's a good segue actually to hearing from Julie, because she used to work for a company doing user research and customer development, but she recently went independent and actually has a startup that creates personalized meditations for children. Such a great idea for a company, right? And needed to create an interactive website to do that.

And she wasn't sure how to do it. She wasn't a web developer. She didn't have these skills. And what she did, to our point about who do you turn to, do you need actual help, is she created a fake team, a multi-agent team that could help her do it. And it worked. Let's hear it.

We can talk about what this makes us feel, but than to help her do it. And it worked. Let's hear it. We can talk about what this makes us feel, but let's hear her describe it. So I went over to Claude and I said, you are managing my development team. On my team, I have a software engineer, I have a code executor, I have a designer, and I have a planner. I'm going to tell you what I want and I want you to manage these four people for me and solve my problem." And it was bonkers. Over the course of the next few weeks, I completely, from somebody who doesn't know how to code,

it's not just not even knowing how to code. It's not even knowing how do you use GitHub? How do you push to a server? What's my tech stack? How am I doing any of this? It wrote me every single piece of code that I needed, told me where to put it, told me how to execute it, helped me get set up with GitHub. I got Google Analytics.

I mean, this was a website I wanted to look a certain way. I can't just copy and paste every single thing. So it would create some code for me, and then I would ask it questions like, why did you do this? Explain this to me? How am I using this technology? Why are things working in this way? And so I will never be a software engineer, but I am a lot closer than I ever was. And like now I can actually have conversations that I couldn't have before. And I was honest with it, which more honest than I would be with a real person.

I would say, look, I don't understand this. Explain it to me like a person who doesn't understand this. But it was bonkers. I mean, it worked. I don't understand this. Explain it to me like a person who doesn't understand this. But it was bonkers. I mean, it worked. I did something I could never do. Well, and that sound you just heard was another industry going down, right? The death knell, the death knell.

But to our point earlier, I do think there's something missing. I don't know if it's a lack of expertise, but that sort of creative energy. But then I hear in Julie's voice, it sounds like she got it. I don't hear anything sort of lamenting the loss of human interaction. Well, it definitely addressed her need and did it in an efficient way, in a way she couldn't have done on her own, to me it highlights the importance of the smart dumb question, you know, the generalist

intelligence, because we're blinded often by our expertise. And in this case, a lot of what she brought to the party was knitting together all of that work by asking the non-expert questions, you know? Yeah. My eighth grade teacher, eighth grade math teacher, had this big sign in her classroom that said, the only dumb question is the one you don't ask. And I think about that a lot in the work that I do, because oftentimes people don't ask the question because they're afraid of looking stupid.

And Julie refers to that. She was asking questions that she probably wouldn't have asked a live human because they're afraid of looking stupid. And Julie refers to that. She was asking it questions that she probably wouldn't have asked a live human because she was afraid of looking like she didn't know what she was doing. Right. It removes some of that ego too, from the process of, I can ask it a really silly question and not have to worry about what it thinks of me. – laughter –

Which actually is a big part of work. Right? Like how often are we thinking, should I say that? Should I ask that? What will that person think of me? You know, like that's what... Oh, for me, it's always, did they already say this? Because I'm so often... Yes. Right. I have started starting to send this. Someone may have already said this, but... Someone may have already said this, but. Yeah. Yeah. Harvard Business School Executive Education develops leaders who make a difference in the world.

Their renowned faculty members are skilled educators. They combine real world experience and a variety of distinctive teaching approaches to create an exceptional learning environment. In their programs, you'll experience the power of fresh perspectives and connect with a world of new ideas. Learn more at hbs.me-learn. That's hbs.me-learn. So Julie, who we've heard from, has been working with small groups, trying to help

them understand what LLMs can do for them. And she poses it as what problems can LLMs solve. And the one that I think probably is most surprising. People had no idea you could solve interpersonal conflict, whether it's work related, whether it's family related, all sorts of stuff. And just the way it approaches solving it was very surprising for people.

So you describe that person that you're having a problem with into the, I use ChatGPT, you could use Cloud. So you describe the problem, explain the whole situation, explain the backstory. I keep having this conversation over and over and I'm really frustrated. I don't know where the actual problem lies and I really wanna fix this. Ask me questions one at a time

to make sure I'm seeing it clearly. By the way, you're an expert at interpersonal dynamics in the workplace. Actually, that's something I might do, is I might say, you're an expert. So before you even describe your situation, you're an expert at handling interpersonal dynamics in the office. When two people are having an argument, you come in all the time and you help

people who have the same conversation over and over, and you often find they're seeing it wrong. Give me seven ways that you have helped people in the past and give me your backstory so I know why I should trust you. And having it go into that much depth before you ask your question really helps it be smarter. All right. So what Julie just described was an Amy G bot. And now I have to ask human Amy G what she thinks of that.

What do I think about 45 different things at the same time. One of the obvious ones is, yep, I'll probably be out of a job, especially if I can train an abot to do exactly what I do, which is to give that advice that she's prompting it to give. So in some ways, it's a thread. I'm putting that in air quotes, to what I do, but I think more than anything I actually find it reassuring because the person is pausing to ask the question, how do I do this differently?

And that is usually anywhere from half to like 80% of the battle in interpersonal conflict, which is to pause, think about what am I doing, how do I not respond from a place of reactivity, how do I respond in a thoughtful way and how do I respond in a way that considers who the other person is and what might matter to them. So the fact that we have tools to accelerate that process makes me so excited. The issue is at some point I envision, right, you and I get into a disagreement on Slack and we say let's just let our bots hash this out. These better evolved versions of ourselves that won't get emotionally worked up who

can maybe just solve the issue. I prefer the trial by combat. Well, you and I like a good fight, let's be honest. But it's forcing us to do the thing that's so hard as humans to do. We're almost hacking our broken brains in those moments of conflict to get to a better answer. And I do think there should be an A.M.E.G. bot. Well, in fact, I'm not quite there, but I'm taking an initial step of creating and training a custom GPT to be me. And actually, with the help of Alex Samuel, who I think you probably know, she's a data

journalist, she's been a long-time contributor to HBR, she's an expert on the digital workplace, and she and I had a conversation about how I could create a bot that's going to help me start to create content in a different way, in a way that's easier and that helps me with the like, wait, what have I said about this before? I know I've said something about this. When do I think of this again? And so I'm not going to be at the point of like a stranger can ask Amy G. But Amy G. is going to ask Amy G. what should I think about this?

What have I said about this in the past? So the next thing you'll hear is my conversation with Alex as she walks me through what all of this involves and helps me get over my big hesitations around it. I'm looking forward to this. Or not. I went through some of your stuff because I was thinking like what would we scrape of Amy's to be able to make a virtual Amy and well Alex you and I have been writing for HBR around the same time and pretty much every time you come out

and say you got to try this new thing and your HBR writing and I go not for me and then like two months later I'm like oh gosh I got to go back and find that and say, you gotta try this new thing, and you're HBR writing, and I go, not for me. And then like two months later, I'm like, oh gosh, I gotta go back and find that Alex article. Like you have always been making me uncomfortable about what I need to do next, and then helping me through the process. Thank you, I really love that.

Also, I think that's gonna be my new LinkedIn byline, making you uncomfortable since 2009. That's right. I'll be honest, before we even started talking about this episode, there's part of me that really wants to create a Amy GPT that can give advice. Like that's what I'm in the business of doing. And so I'm not ready to be honest for a variety of reasons to do that. But I think ultimately I do want to get there, but there's some serious hurdles

mentally for me to get there. So I think by trying to make an internal GPT, I think it'll help me build more trust with the technology and get me more comfortable. One of the things I want my GPT to be able to do is help me with brain fog, which is related to menopause certainly, but also related to age, also related to overwhelm, occasional burnout I have, right? So I actually have written so much,

I don't even know what I've said on certain topics. And so that's part of what I want the GPT to help me do. I also want it to help me come up with new ideas, like make connections between things I have written or said or posted about and come up with new ideas for newsletters or articles. And just sort of be the better Amy. Like, can it do that? So even when you customize an AI with your own content and with thoughtful instructions that reflect how you like to work, it is not

you. And that's good for those of us who still like to earn a living by being ourselves. And while as writers and thinkers, we're always interested in what's new and what's next, for our readers and the audiences and the organizations we are trying to serve, often what is most useful is what we have already shared, already thought through, already said, but updated to this context, a new framework, a new structure, a new social network. And that is where these AIs are just terrific, is kind of taking what you have already done and remixing it a thousand ways in two minutes. Right.

Okay. Can I actually trust what it's going to do for me? And can I trust it to not make me look dumb? I don't think I've published a sentence that was written by an AI except in instances where what I'm doing is saying here's something that was written by an AI but I also can't think of when I last wrote something without the help of an AI and there are a lot of mechanisms you can use to improve the quality of what it's doing for you in the way of background

research, including telling it what to draw from. And that's what you do when you give it a collection of your own work. And that's what you do if you ask it to summarize 10 specific PDFs rather than go and you know find its own sources on workplace conflict. So you know in my case for example I have this Claude project I have it both as a Claude project and as a custom GPT that I call the Alexerizer and the purpose of the Alexerizer is to make content that sounds like me, Alex. And in order for the Alexerizer to make content that sounds like Alex, it has to understand what Alex sounds like.

So I have given it a whole bunch of files that are kind of a snapshot of the breadth of my work in different contexts. And by giving it that range of content, it is able not only to sound more like me, but it understands my work, it understands what kinds of work I've done over many years, and it also understands how I write differently in different contexts. So if I want to train the aimerizer, and I'm going to be able to pronounce that better by the end of this episode, if I want to train it on my HBR articles, my LinkedIn posts, and my newsletters, how much of each of those things would you suggest? Well, I would suggest a couple things.

What I would probably do in your circumstance is I would think about kind of your greatest hits or your foundational pieces. Like you're trying to configure it with three related sets of knowledge. One is actual stuff you think and know and say. So you need to make sure it has the core knowledge, your core principles, you know, the things that you come back to time and again as core themes, you want to make sure those are all in there. So that's the knowledge piece of it. Then you want it to understand style. So from that point of view doesn't necessarily take a whole lot, right? You want to pick the newsletters that really reflect what you like most. And again, you can have a separate file

that is the voice file, and I've done that as well. I've picked, there's the ones that I have that are my body of knowledge, and then I'll pick three or four articles that are, this is gosh, if I could always write this well, this is how I wish I always sounded. And then those are the voice files. And then there's a third piece beyond

just the knowledge and the voice, which is process. And so part of where my Alexerizer provides value is understanding how do we go from Alex's 2000 word first draft to her 1200 word second draft. And because I hate that process, I love it when the AI can help me. And then also, when I finished the article or finished the newsletter, how do I go from here's my 1200 word newsletter, now I want that chunked into a series of LinkedIn posts. So in order to train the AI on the process, I've given it some kind of matched pairs of content. I have an article I wrote for the Wall Street Journal where I gave it every draft along the way, including the email exchanges with my editors and what kind of feedback they gave me. And

then I gave it the final article so it can understand, here's how we go from first draft to published article. And then I did the same thing with one of my newsletters. Here's how we go from first draft to published article. And then I did the same thing with one of my newsletters. Here's a newsletter, here's a series of LinkedIn posts based on that newsletter. I want you to use that as a model. And that's what goes in the custom instructions. When you're taking Alex's newsletter and turning it into LinkedIn posts, use these two files and use that as your model. So what I would say is in terms of volume, I would think in terms

of like those use cases, what do you want the AI to do for you? Do you want it to be writing your articles for you? Probably not. Do you want it to understand how to turn your notes into a first draft? That's what I want it to do. So the next time you're writing an article from notes and interviews, take the interview transcripts. If you use a recording tool that does AI transcription, take the transcript. Then take the notes where you have personally picked out the quotes that you thought were useful from those interviews. Then take your outline, take your graphs, put all those pieces together into a file called, you know, you might have one file called interviews for article, one file called notes for article, and one file

that is the final article. The more you kind of annotate and structure that notes file, the easier it is for the AI. So say, you know, outline first draft, you know, write these as subheads within your notes file. And then you give those to the AI and you say, this is your model. Take a look at how I choose the quotes to use. Take a look at how I turn an outline into a draft. You're teaching it to think like you by giving it examples of you along all stages of the process. Yep. And then giving it examples of you along all stages of the process. And then create the greatest hits file that is your kind of content file that gives it, here's the essential inside of Amy's brain.

Give it your bio, your speaker profile bio, so it knows how you position yourself in the world and then give it some samples of voice and you can even annotate those to say, this is my voice on HBR, this is my voice as a speaker, this is my voice on LinkedIn. And when you write the custom instructions, which is a separate file that can be quite long, you know, I think on GBT the limits 8,000 characters, that's where you tell it how to reference these different pieces and how you give it the overall instructions of what it's doing for you.

Namely, writing in your voice, writing first drafts, assimilating background material. And can you change those instructions? Anytime. Okay, good. I change them all the time. What does the future hold for business? Can someone please invent a crystal ball? Until then, over 40,000 businesses have future-proofed their business with

NetSuite by Oracle, the number one cloud ERP, bringing accounting, financial management, inventory, and HR into one platform. With real-time insights and forecasting, you're able to peer into the future and seize new opportunities. Download the CFO's guide to AI and machine learning for free at netsuite.com slash womenatwork. That's netsuite.com slash womenatwork. Okay, one of the things we haven't talked about, which I think is really important, is getting permission to do what I'm doing from whoever needs to actually know.

So in my case, I'm not a full-time employee at HBR, but because of this podcast, because of my writing, I am subject to their Gen AI acceptable use policy. So that's a piece we've started to figure out. And actually, as we've been talking, we got a note from the compliance and data associate director saying that we could actually use Claude for this, which is great. So I just want to note that for anyone following along who's like, I'm going to do this too, right? Important check mark.

What kind of advice do you give people in terms of getting, making sure they're complying with whatever appropriate rules, laws, policies? There's two things. One is, and this is perhaps a slightly sensitive issue, always hold onto your copyright. We are all generating content all the time. We are creating the seeds for our own self-replacement. If you are an employee and your work is therefore work for hire, then everything you type between nine and five belongs to your employer. And that's right.

So I'm more worried about it, to be honest, from the employee's perspective than I am about the employee violating compliance rules. But for sure, you want to keep on top of your employer's policies, and I think you also want to be really careful about where you create these files. So yes, there is the theoretical possibility that your employer can take all your work product and make a virtual you, but most employers aren't there. But if you have meanwhile taken all your own work product and concatenated it into these giant files and built your own custom version of you that you're using and you have done

that and stored those files on the company file server, you're sure making it easy for them. So I would actually recommend to people that while you obviously don't want to violate your company's policies on AI, you also want to protect yourself by making sure as much as possible that when you're creating these tools that allow you to like automate portions of your own workflow, that those tools are on your computer, that those files are on your computer, again, as you know, within the bounds of what's consistent with your company's policy because we should all be worried about protecting

what remains of our own personal IP. Yeah so while we stay up at night wearing that our employer is gonna replace us with Gen. AI we also want to make sure we don't get fired before that even happens by violating the policy. That's sort of how I interpret what you just said. I think that's a good way of putting it. Yeah. So I think I have what I need to get started, but I'm curious, what didn't I ask you that I need to know just from like a practical step, I know I need to

think about the use cases, I need to think about voice, knowledge, process, files, the instruction files. It's a lot of work with hopefully the result that it's going to save me time. But what else haven't I asked? One thing I would say is really helpful to keep in mind when you are writing those custom instructions is to tell it both what you want it to do and what not to do. Let it know what your weak spots are. Say you'll see in my example that I use

a crazy amount of M dashes, but everybody's always telling me not to use so many M dashes, so your drafts shouldn't use those. Like even small things. Word repetition. Yeah. Yep. Oh, here's a real one. It is a huge challenge to get an AI to write a conclusion that doesn't begin in conclusion. They love to begin in conclusion. So actually, I, part of my custom instructions and some of my AIs is never say in conclusion, just don't. I love that. I love that. So wait, how do I know when I got it right? I could just seed my AI system, my custom GPT with like knowledge files and tweak the instruction. Like I could just do that forever. How do I know that I got an A plus? That's what I'm asking Alex.

There is no A plus. I'm sorry, Amy, you will never get an A plus. The finish line will move infinitely. I mean, on the one hand, I hate all the AI hype. And on the other hand, I basically want to tell you that I think this is the biggest revolution in how we think and connect and work as people. That quite apart from the tangible benefit of

getting your amyarizer working so that you can make your articles faster. That process is how you're gonna understand what is happening to our world now, what's happening to our work and what can happen to each of our own creative processes and it is terrifying and overwhelming all the time. And in a way, that's exactly why I want like you personally to be doing this is because the nature of your work is such that you are one of the people who needs to help everybody else through all of the rough moments that are going to happen during this transition. Okay, I'm excited and terrified.

And that was like an incredibly good pep talk. So thank you. I'm ready. Like not only, not only can I do it, not only like will I do it, but I need to do it, you just told me, which is okay. Here we go. So Amy G., it's been a couple weeks since you mentioned that you are programming the Amy G. bot. How's it going?

It's going well, actually. I would say the interesting thing about the process is that it was a huge hurdle, mental hurdle for me to overcome. And I knew that. In my conversation with Alex, I knew I was going to struggle with finally taking the step to do this. And it was even bigger than I expected. And our poor producer, Amanda, kept messaging me, have you started? How's it going? And I was very honest with her and told her, I haven't done it yet, I'm gonna do it, I'm gonna do it. And the hurdle was really about sort of trying to figure out what is this thing doing? And is the work to set it up going to be worth it?

So what does the AmyBot do exactly? So right now, it's really to help be a partner in my thinking particularly around like my newsletters, my LinkedIn posts, maybe even article ideas. It's meant to help me come up with ideas and maybe copy for short things like posts not obviously an article that just sort of get me started. And so it's supposed to save me time. One of the things I know you can appreciate this, and I'm sure many of our listeners can, which is that one of the things that has suffered in the current busyness of my schedule is time to think. And so just

having a partner who I can bounce things off of is what I was looking for. Okay. So give us an example of how AmyBot helped real Amy think. Yeah. So, this is the thing it's been most helpful at. I have a newsletter for my own business that I send out twice a month. I uploaded every newsletter I've ever written into it and said, can you help me think of ideas for future newsletters? And it came back, because I had given in the instructions, I said, don't answer the question right away, ask follow-up questions, which I think I might

need to change that instruction, but I'll explain in a minute. So it asked me, do you want me to look at fresh new ideas? Do you want me to look at follow-ups to existing articles? It prompted, what do you actually want? And I said, oh, I'd like a follow-up to an existing newsletter. And it pulled one from January 2023 and said, you mentioned gossip in this article. Maybe you want to write a newsletter all about gossip. And it gave me three ideas, all of which were great. And I will say the reason I'm hesitating on the instructions was because that instruction to ask follow-up questions meant like, you know, so many people talk about how AI is like a smart intern

It didn't feel like that because I was like the intern needs to stop asking questions and just give me an answer right and at One point I would say something it would ask a follow-up question Or three questions, and I say can I answer those one at a time? And they'd say yes, and then they'd already have a follow-up question to my first answer. And I'm like, wait, we're losing track of our question. So it needs to be more discerning in its follow-up. Yeah. I think I need to instruct it to produce more and ask less.

So it helps you generate ideas. What else does it do for you? So I named it Esme, because that's really what I wanted to name my daughter. And I couldn't, because we have a friend who had just named their daughter Esme. So I decided I named it Esme because that's really what I wanted to name my daughter and I couldn't because we have a friend who had just named their daughter Esme so I decided to name it Esme. So I talked to it as Esme and this is the sort of high level instructions I gave it.

You support Amy in drafting social media posts, coming up with newsletter topics, producing rough drafts of newsletters, reminding me of advice I've given before, helping me respond to people who ask me for advice and more. It's a lot I'm asking it to do. And I think I will hone it a little bit more and like maybe create just one that's focused on newsletters or maybe one that's focused on LinkedIn posts or maybe one that's focused on advice giving. Or I sort of need to play with it more,

maybe I just have different conversations with it that do each of those things. This is where I'm a little lost, like in the process, and I probably will go back to Alex and say, okay, I got pretty far, I'm really happy with what I've done so far, how do I tweak this? So what did you train Esme on?

Okay, I have a couple variations on my bio, so I gave her all my bios so she could understand who I am. I gave her a summary of both of my books. So I didn't give her the whole book. We're using Claude AI for this, and I think it would have taken up too much of the knowledge capacity is what they call it. I'm already at like 50 something percent of the knowledge capacity. So I gave her summaries of both my books.

I gave her transcripts of our episodes I did about getting along based on my book, partly because our producer, Amanda, already had those. So I was like, okay, that's easy to get, I'll just put those in. And then I did all of my newsletters that I've ever written. And then I did a sampling of some LinkedIn posts because I couldn't get the full history.

And then I actually, last night I was talking to her about, I wanted her to help me write a LinkedIn post on an article I published on Monday. So I cut and paste the copy from that article as well and put that into a text doc for her to look at. She didn't do a great job with that, I have to say. Like it's good, but especially for writing, like once you start to really look at. She didn't do a great job with that, I have to say. Like, it's good, but especially for writing, like once you start to really look at it, it's very surface level. So I wonder if you've had the same experience that I've had where you've, you ask it to

write something and you ask Esme to write something. And usually for me when I'm doing it, my state of mind is usually anxious and sort of a little guilty that I'm doing this and also sort of kicking myself because, God, why did I wait till the last minute again? Yes. So that's usually the whirl of stuff that's in my head. And so when Claude spits back something, I look at it very quickly.

And my first response is, phew, I can work with this. And all of the noise goes away, which gets me to the point where I can look at the draft for real, and I see, oh, no, no, no, no, no, no. But at least it's something I can work on. And I think I almost use it in some ways of what I don't want to do. Like, I do find the idea generation really interesting. So when I asked Esme to produce a few newsletter ideas, I loved the ideas.

But I didn't ask her to write about them. But it just feels very, I don't know, especially as editors, I'm like, this is like that jargon piece from a writer where I thought at first glance it was going to be great and then once I started editing it, it all fell apart. I realized I need to rewrite the whole thing. Yeah, yeah, that's painful. I mean, for me, writing, one of the reasons I like Claude or Copilot to give me a draft to work with is that it's much easier for me to edit than it is to write. Writing can be torture, but when you're really in it, it's thinking.

It's thinking. And it's figuring out what you think. Yeah. And I think I like writing more than I like editing. I wish I trusted Esme, and I haven't tried, but I wish I trusted her to edit, because that would be fun. But I... I wonder what Esme would say to you if you said, Esme, I know you want to edit, but I don't trust you yet.

I kind of wonder what she would say, because she already has at least some of your thought process in her. Yes, I can do that. I can do that right now. Here's the other thing that's funny. I have to refer to her by name when When I start typing I go, Esme. And you have to be polite. I'm so polite to her. And I watched the emotional labor I do to make sure her feelings aren't

hurt when she produces something crappy is amazing. Maybe she's a machine, but go ahead. I do the same thing, but I do think it's worth reminding ourselves. Okay, here's what I'm writing. Give her the unvarnished truth. Okay Esme, I would really like your help editing LinkedIn posts but I don't quite trust you as an editor. She's thinking, she's thinking. She's taking a deep breath. Yeah. That was a gut punch. I know. Well it's funny when she's thinking it says ruminating on it. Yes.

No, she asked a few clarifying questions. Oh my god. Yeah, she wants to help. She says, please, let me better understand how I can help you with LinkedIn posts while respecting your hesitation about editing. And then she asks, what aspects of LinkedIn post editing make you most nervous about delegating?

Oh my god, just like talking to your therapist. What elements of your voice and style on LinkedIn feel most important to preserve? Would you prefer to start with me reviewing existing posts and providing feedback rather than editing directly? Oh, I think she just gave you a path forward. She did. I have to say, I feel like we are going to end up in a long conversation that I'm like, I still don't trust you. I think you should thank her and close your laptop. With an exclamation to X. I'm gonna go. I'm just gonna do thank you Let's see what she does under the exclamation point to X. I use to exclamation point. Oh, no, she has more questions Sure, she does as may as may really needs to just back off. Yeah

All right. Well, Amy, this has been so interesting and I'm so curious about where you take Esme. I will keep you posted. Please do and please say goodbye to Esme for me. I didn't mean it. I swear. I'm just closing my laptop because I can. Bye. Before we end this episode, there's one more person we want you to hear from, especially

those of you who are self-employed. She's figured out how to use Gen.A.I. to make growing her business just a little easier. My name is Teresa Ramos. I work as an executive coach, a technology consultant, and trainer. I love my job, but a big part of it is writing proposals, which I absolutely hate. A client approaches me and we hit it off, we think it's great. And then they say, okay, could you write a proposal? And I'm like, oh no. And they would go like, oh, you know, you just have to write what you just told me. And I'm like, yeah, right. You know, because I dreaded just a request would take me maybe a couple of weeks. So one day I thought, how about getting

chat GPT to do it? Let's go through the process. I go and talk to the client. And when I go and talk to the client, I take notes. In this case, I have my own GPT for writing proposals. In this case, I have my own GPT for writing proposals. And in the instructions, you can say, act as a strategic director, act as a professional copywriter, whatever, act as a very experienced technology consultant. You're creating proposals. You give a description, a generic description,

because remember, this is something that you're going to use for all your proposals. proposals, you give a generic description, because remember, this is something that you're going to use for all your proposals. And then to make it even better, you can upload previous proposals that you have already created. So now, ChatGPT knows who it is, and what the proposals look like, and how they are structured. So whenever I need to write a new proposal, I just say,

this time you're giving a one and a half hour training session for a group of 30 senior executives. Please write a proposal. The proposal should include a description of the session, learning objectives, and learning outcomes. And then you hit return and then it will write you the proposal. And not only that, those 20 minutes or that half an hour, it's not like pulling teeth out because it's like, oh, let's see what comes out now. Women at Work's editorial and production team is Amanda Kersey, Maureen Hoke, Tina Toby-Mack, Rob Eckhart, Erica Truxler, Ian Fox, and Hannah Bates. Rob and Maure compose this theme music.

I'm Amy Bernstein. Get in touch with me and Amy G. by emailing womenatworkathbr.org.