How I AIChatGPT agent mode: The “little helper” that transformed recruiting & solved parking nightmares
EVERY SPOKEN WORD
45 min read · 9,300 words- 0:00 – 2:10
Introduction to Michal and ChatGPT agent mode
- CVClaire Vo
We're gonna start with something that we haven't actually seen on How I AI yet, which is agent mode in ChatGPT.
- MPMichal Peled
My use case was with our hiring team. Part of their workflow is to browse through many LinkedIn profile and search for relevant candidates. It takes a lot of time.
- CVClaire Vo
Let's talk about the prompt. I'd love for you to go through how you thought about structuring it to make it effective with the agent.
- MPMichal Peled
I want a little helper. I'm a recruiter. I want someone who is like me. So I started by telling it, "You're an IT recruiter," and then I described what I want it to do.
- CVClaire Vo
I love that you called it your little helper, because don't we all want an AI little helper? [upbeat music] Welcome back to How I AI. I'm Claire Vo, product leader and AI obsessive, here on a mission to help you build better with these new tools. Today, I have Michal Peled from HoneyBook, their Technical Operations Engineer, who's building tons of internal tools and automations to make their team's life easier and reduce friction. Michal's gonna show us some advanced features of ChatGPT, including agent mode, replicate not one, but five of their personas as AI identities, and save me a lot of time on my commute using ChatGPT. I'm really excited about this episode. Let's get to it. This episode is brought to you by Brex. If you're listening to this show, you already know AI is changing how we work in real, practical ways. Brex is bringing that same power to finance. Brex is the intelligent finance platform built for founders. With autonomous agents running in the background, your finance stack basically runs itself. Cards are issued, expenses are filed, and fraud is stopped in real time without you having to think about it. Add Brex's banking solution with a high-yield treasury account, and you've got a system that helps you spend smarter, move faster, and scale with confidence. One in three startups in the US already runs on Brex. You can, too, at brex.com/howiai.
- 2:10 – 5:14
Using agent mode for LinkedIn recruiting automation
- CVClaire Vo
Michal, thank you so much for joining How I AI. I'm excited to see what you have to share.
- MPMichal Peled
Thank you so much for having me.
- CVClaire Vo
We're gonna start with something that we haven't actually seen on How I AI yet, which is agent mode in ChatGPT. And so I'm wondering if you can just go ahead and dive into what was the problem that you were trying to solve, and why was this agent mode, this agentic browsing, the solution to the problem you were having?
- MPMichal Peled
Our problem was, um, you know, same as, same as our customers are having. Uh, you have to do your job. You have a job that you really love doing, and you have your proficiencies and, uh, and expertise. However, you spend a lot of your time doing the, the, um, mundane, thoughtless, uh, manual, repeating work in order to do... uh, to get the information that you need. So my use case was with our hiring team, and as a recruiter, when you get a job description that you need to recruit to find candidates for, part of their, part of their, uh, workflow is to, uh, browse through, uh, many LinkedIn profile and search for, uh, relevant candidates that may be, uh, relevant for the job descriptions, and it takes a lot of time. It can be hours of browsing through profiles and going through all of the characteristics that they're looking for. So I wanted to take that load off of them, and, uh, ChatGPT agent mode came just in time. We all talk about what agent is and what agents do and how we can use them. In ChatGPT, it's very, uh, simple to understand. So you just open a chat with ChatGPT, but then you add an instruction and turn it into an agent mode very simply from the toolbar. And once it goes into agent mode, it means that it can take the prompt, or you can actually use specific prompts, to tell it not just to search for information online, but also to perform actions for you. And why did I need it in this case? Because I needed to log in to LinkedIn. I don't want it to just search for profiles, uh, on LinkedIn, just, just profiles that are publicly accessible. That's not the information that I need. So I needed it to log in into LinkedIn, and I needed it to perform search, and I needed it to go through the profiles and, and look for the restrictions that I wanna give it, and those restrictions were provided by the actual, uh, hiring team, that they actually use it as, uh, requirements for potential
- 5:14 – 10:50
Creating effective prompts for agent mode
- MPMichal Peled
candidates that they find.
- CVClaire Vo
Yeah, let's talk about the, the prompt really quickly, because I think this prompt is, is interesting as I'm reading it, and I'd love for you to go through how you thought about structuring it to make it effective with the agent.
- MPMichal Peled
Of course. So I usually start my prompts, uh, begin my prompts with telling the GPT its role, and so here I told it, "You are an IT recruiter." I want a little helper, right? I'm a recruiter. I want someone who is like me that will h- will assist me with my job. So I started by telling it, "You're an IT recruiter," and then I described what I want it to do, what the task is: "Log in to LinkedIn using my account. If not already logged in, let me take control and log in." It is something that is possible.... find up to five LinkedIn profiles where the current title and job description match the attached job description. Uh, and here is the part where I just uploaded a job description. In this case, it's for an engineering role, okay? Um, so I have the job description, and I have the, um, um, "You are an IT recruiter. This is your job. This is the task," and I provide, like, a full description of the task, actually, actually describing what an actual IT recruiter would do. And then I added restrictions or special instructions. It doesn't matter, um, matter how you call them, but these are important because these give- don't just search for something that matches the description. Do it the way that we do it. And when our hiring team goes into a search, they have specific criteria that they go for. So I collected these, and, and I added it as a list, as a restriction. I could call it instructions. It would, would've been the same. So candidates must be from Israel because the, the job is being filled up in Israel or currently working at an Israeli company, and they must be active in LinkedIn, uh, within the last three months, because that's something that our hiring team is looking for. Uh, and the current job, job role must be close enough to the, uh, open role and in title and seniority, and, uh, also something that is special, uh, the candidates must either work in their current workspace more than a year, or they can be unemployed, but no more than a year, uh, and have worked in their last workplace for over a year. These are all things that I... I didn't invent them. They were taken specifically from our hiring process.
- CVClaire Vo
What I love about this is exactly what you said, which is, uh, first, I love that you called it your little helper, because don't we all want an AI little helper? That is, that is my goal. Maybe I'll rebrand my product to just Little Helper. Um, but what I like about this is, you know, it- when you're building a, a tool like this or a prompt like this, the simplest way to get to a good outcome is simply interview somebody and say, "Step by step, just tell me what you do. Like, tell me what you do." And if you can codify what a person's step-by-step workflow is, and you can put that into a pretty simple prompt, which here it's only a paragraph and, you know, three or four bullet points, you can replicate and automate that at scale. And typically, this is not the highest order of thinking you want your recruiter or sourcer to do. You don't want them just to build a list and be looking, "Is this person here a year or not?" That is an input to what you hope is a great recruiting process, great outreach, all that kind of stuff. So, one, I think it's just really great to interview your colleagues and say, "How do you do your job, and what parts do you hate? And let me automate them." The second piece that I think is really interesting here is you're actually pretty specific about a couple outcomes. You're specific about the number of candidates that you want, so I think that's really helpful. You're specific about a kind of, um, threshold of matching your criteria, so you say seventy percent, and I often find these LLMs are very... You know, what is seventy percent matching? Like, it's not, [chuckles] it's not pure science, but they're pretty good at, at, at following a general threshold, so I think that is really interesting. And then the last piece I wanna call out, which we'll maybe see in the demo, [lips smack] is agent, while it can be agentic and independent, can also be a co-pilot and collaborator. And so you actually instruct the agent when you're gonna take over and when they're gonna take over, and so I think that is also really interesting things for folks to know, is you don't just have to, like, press the button and walk away and let the agent run. You can press the button and say, "Hey, wait. When you get to this point, let me take the next step," and then you can go on from there. So that's really interesting.
- MPMichal Peled
Exactly. Or if you encounter a problem with this and that, stop and ask for my assistant. Um, and that's exactly the agent, that's exactly the agent mode. Thinking about it as a, a little helper will really help you come up with, uh, good prompts for it.
- CVClaire Vo
O- okay, I think you gave us our-
- MPMichal Peled
Yeah, and-
- CVClaire Vo
... and our, our show
- 10:50 – 16:29
Demo of agent mode searching LinkedIn profiles
- CVClaire Vo
title will be ChatGPT agent mode: Your Little Helper. [chuckles] So let's see how it runs.
- MPMichal Peled
Exactly. Exactly.
- CVClaire Vo
[chuckles]
- MPMichal Peled
Um, so o- once you start running it, and it's- this is something that is mind-blowing for anyone who, who, who tries it for the first time, even the ones who are, you know, very proficient with using AI tools. Suddenly you get, like, a computer. [chuckles] Open it up in, [chuckles] in your, you know... You see, you, you see your, uh, your, your little helper actually doing things in the computer. So it start- it started by reading my job description, and then you can see it goes... It, it, it will try to go to LinkedIn. It will probably be already logged in because I logged in beforehand. And, uh, the thing I like the most, you see, is it's logged in, and then you can see, like, the arrow goes and clicks on things and searches on things and go through the list. And the thing I like the most is, is that during all of this time, you can see the thoughts of the agent. "Now I will go to, uh, the feed page loaded again. I plan to click on... First, I need to make sure," like, you can see inside the brain of your agent while it is thinking.... And so, uh, and all of this is live. So I will let it run here, but we can go and see, uh, results.
- CVClaire Vo
Yeah. You know, I, I wanna call out a couple things, because I know that often on How I AI, we have highly technical use cases for highly technical folks, but we also have a lot of people that are actually quite new to using generative AI tools and have probably been pretty familiar with ChatGPT and the direct experience. But I know if I showed my mom this, or [chuckles] um, even some of my friends that maybe don't work in tech, and said, "Hey, did you know that ChatGPT can open up a magic computer, and navigate it, and narrate its thoughts, and look for things for you?" They'd be pretty, pretty surprised. And I think, you know, as you watch this, uh, what I hope our listeners and viewers are taking away is, you don't just have to rely on text prompts and chats when you're using these tools. Now that the next kind of, like, evolution of these LLMs, especially the more like consumer-focused, general-purpose ones like ChatGPT, have evolved, you actually have a lot more tools. And so I just wanna call out, so for f- some of those folks, um, out there, I'm... You know, I'm thinking a little bit of, of my parents and some people in an older generation who are like, "How do I get from here to here?" Or, "I need help, you know, searching for flights," or, "I need to do a certain kind of research on a niche site." Having sort of this expert computer operator on hand, I actually think is gonna make, um, information more accessible to folks, but it's also going to make UX and websites more accessible to folks that don't have the time to figure out, "How do I use the best filters on LinkedIn?" or those sorts of things. And so I just- I wanna make sure that people that have not experienced agent mode, and I know we're all on the edge, so maybe all of us have. Just, you know, take a minute to appreciate the kind of use cases that this opens up. Also, it's just fascinating to watch-
- MPMichal Peled
Yeah
- CVClaire Vo
... AI operate-
- MPMichal Peled
It is, it is.
- CVClaire Vo
Sure. Yeah.
- MPMichal Peled
Of course, first time I, I, I tried it, I just sat and watched [chuckles] the thoughts of the agent while it was thinking.
- CVClaire Vo
Going to our more technical audience, though, a couple things that I wanna call out is, one, props to the OpenAI ChatGPT team. What a great user experience design here. Like, it could feel very strange to watch, uh, an agent browse a website. It could either be boring or weird, and I think this, um, user experience of, like, being able to see where the cursor is, showing the reasoning and thoughts, watching it navigate, is actually pretty entertaining. That's a hard thing to pull off for a, a consumer product, and so for anybody designing AI products, it's worth thinking about some of these interaction patterns here. And then again, I just think about how long this would take someone to do. We're watching it because I'm trying to narrate some of the features, but, you know, you could, you could walk away, you could go to a meeting while this happens, or you could-
- MPMichal Peled
Exactly
- CVClaire Vo
... do something [chuckles] something else.
- MPMichal Peled
If you have ChatGPT on your phone, you will get notification, um, a notification when it's done to tell you that it's done, and, and here are the results for you. We said little helper, [chuckles] so.
- CVClaire Vo
And the last thing I'll c- I'll call out, I love... I really do love this use case, is I do this specific thing all the time. As a hiring manager, executive leader in an organization, I was constantly looking for, like, "Who's a senior director of DevOps and platform infra who is either in San Francisco or works for a San Francisco-based company, who has experience in dev tools?" That, you know, one, one change I would maybe make on your prompt is, like, is one or two connections away, so I can actually message them or get a, a back-channel reference on that. I did this all the time. I had my hiring managers do this all the time. So even if you're not in recruiting and you're just somebody who does hiring, I think this specific workflow is really, really useful.
- MPMichal Peled
Yeah, definitely. Uh, that's an excellent use case that you, you just mentioned.
- 16:29 – 19:53
Results and team reception of the recruiting automation
- CVClaire Vo
So let's actually, um... Let's look at the output. We'll, we'll let this run, but I know you have an example output for us.
- MPMichal Peled
Yeah, yeah, it runs. I have an output that actually, uh, worked for 10 minutes, just that, and within these 10 minutes, I got a list of five candidates, as I requested for, and you can see the match score. Well, m- having a match score or rank for, a result is something that I really love doing. I- it's not a must, but if you give, uh, specific- if you provide specific requirements and ask for a match score, it is easier to understand what, uh, results are more- have more quality for you. I mean, otherwise, you could just get a table of, like, these are the five results, but is someone of them better or maybe a better match for what I need than the others? I won't be able to see it unless I instruct the, the GPT to provide me with some score. So it's not, um, an exact science, but it does give you some kind... uh, a way to compare between the results. So I will say someone who got 90%, um, match, is probably, uh, as, as, like-- it probably will be a better match than the 78% match, and I will have to go deeper and understand why.
- CVClaire Vo
Yeah. One, I, I'll call out a couple things here that I think are for interesting for people to look like. I know we were talking before the show, you actually made this anonymized data just so we weren't showing people's profiles or, you know-
- MPMichal Peled
Yeah
- CVClaire Vo
... showing how person A versus person B-
- MPMichal Peled
Yeah
- CVClaire Vo
... fit a specific job you're hiring for. But I will say, anonymizing candidate profiles is actually a pretty standard practice in a lot of recruiting flows, just to make sure-... you're not biased, this school, that school, this person, that person-
- MPMichal Peled
Right
- CVClaire Vo
... this name, that name.
- MPMichal Peled
Right.
- CVClaire Vo
And so I actually kinda like this flow, where you're really just comparing the qualifications against your stated objections or objectives. And so I think that is a really interesting kind of meta flow that you're showing here.
- MPMichal Peled
Yeah.
- CVClaire Vo
The second thing that I [chuckles] wanted to say as I was reflecting back on agent mode is it's almost exactly like a recruiter or sourcer would navigate LinkedIn, except for one thing: when I log into N- to LinkedIn, I don't go straight to the job to be d- I don't go straight to that search bar and search for, like, VPs of engineering.
- MPMichal Peled
Yeah.
- CVClaire Vo
No, no.
- MPMichal Peled
You get distracted. [chuckles]
- CVClaire Vo
I get distracted by the, by the notification. I start reading the feed.
- MPMichal Peled
Yeah.
- CVClaire Vo
I'm responding to comments. I go through my inbox. And so I think, like, why is it 10 minutes? It's 10 minutes because it's, like, pretty hyper, hyper-focused and efficient, but it's also 10 minutes 'cause you're [chuckles] not getting so distracted with all the other things in, in the application, and you can really just get the agent to focus on the task, task at hand.
- MPMichal Peled
Yeah.
- CVClaire Vo
So, um, maybe it's a way for us to all break... I, sor- sorry, LinkedIn, LinkedIn Growth PMs, I apologize. But [chuckles] it's a good way for us to still get the value of these platforms without getting our time sucked into the, um, less value-generating aspects of them, maybe.
- 19:53 – 23:50
The outcome of implementing on Michal’s team
- MPMichal Peled
Yeah, that is correct. Um-
- CVClaire Vo
So tell me a little bit about how this was received by the team. I'd love to know the kind of outcomes here.
- MPMichal Peled
Yeah, well, I, I, I will be real and say that, um, like, the first result that I got, I was very skeptic about, so I just took the, the table, and I sent it to our hiring manager, and I told her, "This is the job description. This is what AI found for me, uh, in LinkedIn. If you can go through the results and let me... Give me feedback. Are there good results? Are we familiar with these candidates? Did we try to reach out to them, or, uh, you're looking at them and say, 'Oh, no, that's a terrible fit. I don't know why, why, why this person is even in this table?'" Um, and so she went over, over them. You can see that the table as link to the direct, like, the, the LinkedIn profile per candidate. So she scanned those five profile, and she came back to me and she said, "Well, uh, you know what? Out of these five, four of them were never found by us manually, and they really fit the description, so we would wanna a- approach and, and, you know, try, try to get them to, uh, to come for an interview. And the fifth one was actually one that we caught manually and is already coming for an, an, an interview." Uh, and so to me, it was a great sign for quality. I mean, it, it's not just a list of names. Those are actual, real quality candidates that we can work with. And so now, now they want, they want the agent to run on a lot more, uh, job description, many more job descriptions that we have, provide them with more than five candidates. I wanted just five to see if, if it, if it's worth something. Um, yeah, but now it's going to be a real part of, of their hiring process, freeing their, their time to do other things that they love and appreciate a lot more.
- CVClaire Vo
Yeah, and I can't emphasize what you're saying enough because so many people push back on AI, saying, "Yeah, you may get speed, and you may get efficiency, but you're not getting quality." And my experience has, has been the opposite of that. You get speed, and you get quality, and again, it's those- it's that last mile, those edge cases, those ones that are, like, just a little hard to find, a little hard to research, where I think AI can increase the quality, um, of the, of that last bit. And so it was... It's amazing to see that this worked for your recruiting. It's given me so many ideas. Not, you know-
- MPMichal Peled
Yeah
- CVClaire Vo
... not just in recruiting, um, use cases, but just in general, one people finding use cases. I was thinking about how you could find great candidates or customers on, like, X or LinkedIn. And then the other thing that I think is, is really great here is just showing we don't get a lot of, um, GNA functions. We don't get a lot of people functions getting love [chuckles] in how I AI. I feel like all the noise is about, like, product design, engineering support, and so I just love seeing the recruiters get some love here because you're the people that bring in great talent and fun colleagues to work with, so thank you for showing this.
- SPSpeaker
This podcast is supported by Google. Hey, everyone, Shresta here from Google DeepMind. The Gemini 2.5 family of models is now generally available. 2.5 Pro, our most advanced model, is great for reasoning over complex tasks. 2.5 Flash finds the sweet spot between performance and price, and 2.5 Flash Lite is ideal for low-latency, high-volume tasks. Start building in Google AI Studio at ai.dev.
- 23:50 – 28:43
Creating custom GPT personas from customer research
- CVClaire Vo
Let's zip to your second use case, which I think is really... We- we're going from finding real people to creating fake people, so I'm excited- [chuckles]
- MPMichal Peled
Exactly
- CVClaire Vo
... excited about this, this next workflow.
- MPMichal Peled
Wow, that's, that's an excellent description of it. Um, so, uh, let me ask you that. Um, imagine you're a business owner, and imagine being able to talk to, uh, thousands of your potential customers all at once and gather their insights on your planned ad campaigns, planned features, product experience, all from your phone or tablet, 24 hour a day, with, uh, one click of a button. Actually talk to, actually talk to them. I, I, I thought it's mind-blowing, and so it started with, um-... HoneyBook, um, invested in a comprehensive customer research with a third-party provider. We interviewed hundreds of our target small business owners, and they created five detailed buyer personas. But the research was trapped in documents, hundreds of pages of insights that teams rarely referenced because it was too time-consuming to extract actionable information when making product or marketing decisions. So the end goal was, we have five personas, we wanna talk to them, let's create a ChatGPT that is that person, that actual person. And so I started with... And here, the- there are some technical takeaways, but here I wanna put the, the spotlight on the thinking process, because it's very easy to go to ChatGPT, and everyone with a subscription tier of, um, um... It's not the- a plus, a plus subscription tier and above can create, uh, their own custom GPT. And so you go to create a custom ChatGPT. It's quite- it's a quite simple process. You add a name to your GPT, a description, but the most important parts are the instructions you're providing it with and the files that you can upload as a knowledge for that, ah, chat that you're talking to. So I needed five like them. But first I thought, "Okay, this is all I, this is all I know about custom ChatGPTs." I can basically take all of the documentation from the interview and just upload all of the files, text files, presentations, whatever, into this, uh, custom chatbot, provide it with some instructions on, on how to answer, what the research was about, what to say, what not to say, and ask it questions about the research itself. But that's not what I wanted. So I was like: Okay, if I'm taking just the files per persona, so I'll concentrate in one persona, take the files related to that persona, upload it to the chat, instruct the, the, the chat what that persona is, how to read the files, what's included in the files, then I'll be able to maybe ask about that persona, and I will probably get answers like, "That persona would have done this," or, "That persona would prefer that," but it's not like talking to the persona. I'm still talking about the persona and, and not with it. And so I, I, I realized I'm not going to rely on uploaded files for the ChatGPT. I need the instructions to be the, the main and most important part of what consists of that persona. And the instructions here will not be what's in a file or talking about someone else. There will be exact instructions that goes... Th- th- that go like, "You are that person, and this is your belief system, and this is how you run your business, and this is how you deal with media, and, uh, this is how you deal with technology." And everything has to be super, super, super tight. So that ChatGPT, once running live, will actually become that persona that can answer based on what it knows about itself, not files that are attached to it. And so I needed to, uh, bridge that gap. I have all of the research there. I need a, a... I need to make a person out of it, or five, five of them, actually. Okay, so-
- CVClaire Vo
Okay, so you heard it here first. This is our first How I
- 28:43 – 35:00
Using NotebookLM to transform research into persona prompts
- CVClaire Vo
AI, where we are manufacturing not one, not two, not three, not four, but five [chuckles] people. So show us the process.
- MPMichal Peled
Yeah. So I thought about it, and then I decided to go to another tool that I really like. This is NotebookLM. Uh, it's a Google- it's a Google's tool. And NotebookLM, the thing I like about it the most, and the reason I picked that tool in order to, uh, construct the, the, the instructions or the prompt per persona, is, uh... Actually, there are several reasons, but one of them is NotebookLM allows you to upload your own sources and can answer als- only based on these sources, and not things that it goes and finds online, or thinks, or knows, or filling in the gap. This is the information. You ask me about something, I will answer based on the knowledge that you provided alone. Also, it allows you to check and uncheck the, uh, sources that you wanna rely on, so I can ask a question without relying on the buyer journeys, for example, and then the answer will not include that, that part of the knowledge, things that I cannot do in ChatGPT or anything else. Um, and then, uh, there's this chat part within the notebook where it's, it's, uh, it's Gemini-based. Uh, Gemini is this Google, um, chat model. Okay? And, and in this, uh, chat window, I prompted the chat, uh: "Again, you are an expert prompt engineer." Again, with the role, what you are. "Specializing in creating custom G- GPTs by providing strong AI prompts." And then the mission, what your task or mission is. Uh, "Your mission is to create AI prompts for custom GPTs, representing entrepreneurs and small business owners."... who are the decision-makers, and so on and so on. So you will craft highly detailed, nuanced, and authentic ChatGPT prompts for five distinct buyer personas based on your sources. I never told it what were the personas.
- CVClaire Vo
Yep.
- MPMichal Peled
It had to get it from the sources, and that's, uh, the, the most important part, again, is guidelines. I mean, prompts are nice. It, it usually- [chuckles] they usually should come with some guidelines, instructions, um, anything that you want to be, to, to, uh, the chat to take s- specific care of. So in my case, the guidelines were, uh, to ensure that the prompt correctly and fully described the core identity, mindset, decision-making style. I didn't want the chat to decide, uh, on itself what, what I care for about those personas, because I knew what I care for. So I wanted their mindset, and decision-making style, and tone, and communication style, and then, um, the business needs, and the technology stack, and the journey map, social media preferences. I, I pointed the chat to exactly what I needed to get out of this research. And then that's another important one. I think one should not go on without that instruction. "Don't add or modify text that is not written or implied in the text." [chuckles] Okay, I know you're creative. I'm turning you down.
- CVClaire Vo
[chuckles]
- MPMichal Peled
"The text describe a specific persona must remain true to the original persona." Yeah.
- CVClaire Vo
I'm laughing because yesterday I literally wrote a prompt that was like, "Do not make up any links." It... I had a thing that was making up. "Do not make any up, any links that are not in your source of links." Like, and it, and it's so funny, we get so used to operating these chatbots as if they have human reasoning, and sometimes they have kind of like superhuman reasoning, and sometimes they just do stuff that a human would never do, like just make up something.
- MPMichal Peled
Yeah.
- CVClaire Vo
And so I think this third prompt, we'll zoom in on it on the show, uh, is probably applicable to a lot of things, especially when you're trying to constrain, um, an LLM space to a specific set of, of, of data and, and inputs. So it's a, it's a good prompt. Everybody should use the "don't make up stuff" prompt. Your hallucination rate will drop by a significant amount. [chuckles]
- MPMichal Peled
Yeah. [chuckles] Yeah, yeah, exactly. Wow, it, it is very... It, it is crucial-
- CVClaire Vo
Yep
- MPMichal Peled
... to add those things and to think about them, right?
- CVClaire Vo
Yep.
- MPMichal Peled
Yeah. Um, so the result was-
- CVClaire Vo
Yep
- MPMichal Peled
... and, um, another thing that NotebookLM is good at doing is you can, uh, save the responses of the chat as notes, and those-
- CVClaire Vo
Yep
- MPMichal Peled
... notes are saved here for you to look at later.
- CVClaire Vo
Mm-hmm.
- MPMichal Peled
So I can show you, uh, an example of the notes that it created, but mainly you... I just took the prompts. I went over them. Uh, the important thing about the prompts is, uh, that's another, uh, strength of using NotebookLM. It uses, um, it uses citations.
- CVClaire Vo
Mm-hmm.
- MPMichal Peled
So you can actually go over, um, a piece of information that the chat decided this persona is, and see where did you take it from, and, and just make sure and verify that it went through all of your, uh, data information and didn't invent anything. So-
- CVClaire Vo
I'm gonna laugh because I'm from Austin, and I'm pretty sure I know people-pleaser, whatever, Parker. Um, [chuckles] so this seems like a very accurate Austin entrepreneur persona, if folks are wondering if this is creating high-quality, high-quality-
- MPMichal Peled
Yeah
- CVClaire Vo
... persona prompts. [chuckles]
- MPMichal Peled
Yeah, yeah, yeah. So eventually, yeah, I, I- what I did is just
- 35:00 – 37:20
Adding guardrails to custom GPT personas
- MPMichal Peled
I took the prompts. I did need to refine them a little, okay? Because, um, even with all of my instructions, Gemini didn't exactly, uh, realize what kind of prompt it needs to create, so it was missing some guardrails. It was a little too long. Uh, the ChatGPT, the custom GPT, uh, instructions are limited to 8,000 characters, and it created some of the prompts, uh, being longer than that. So I did need to deal- to do a little refine and create, uh, stronger prompts. So, for example, and I'll show you here in the demo, I needed to add... I used ChatGPT itself or sometimes Claude, because I like working with Claude. Um, I used them just to, uh, tighten the, the instructions a little and make it more robust and add guardrails. So I added, for example, um, "You do not act as a general person, pur- purpose assistant. You do not ask follow-up questions. You avoid slang, bad language, or dis- distasteful content, and keep communication respectful and inspiring. You avoid political, religious, gender, or racial commentary." And I really wanted to add it. That's another key point for creating custom GPTs that need to talk as a person, because believe me, those people work with you. They are your friends.
- CVClaire Vo
[chuckles]
- MPMichal Peled
The first thing they will try with a ChatGPT-like persona is to tackle them with swear w- words-
- CVClaire Vo
[chuckles]
- MPMichal Peled
... or their, uh, their, uh, ideas about political things, or, um, I don't know- [chuckles]
- CVClaire Vo
I was gonna say-
- MPMichal Peled
... recipes for, uh, for food, so-
- CVClaire Vo
... this, this maybe should be a default prompt wrapper on all enterprise GPTs, and it would save us all a lot of, a lot of heartache.
- MPMichal Peled
Yeah, so you can see it, uh, it's in order of what I asked for, so cor- core identity, mindset, business needs, uh, technology stack, whatever. And then what you get is, then, then it's the time to actually
- 37:20 – 41:02
Demo of interacting with custom-persona GPTs
- MPMichal Peled
test them. So we created those five, and I can go to Balanced Blake. You can see, uh, she's one of the most talked to [chuckles] internal chats. So we can go to Balanced Blake, and I can ask her, um, "What kind of ad headline would catch your attention," maybe I'll move it, "would catch your attention during a busy workday?" Don't you- wouldn't you wanna know the, that thing about your, uh, uh, ideal or prospect client? And I can an- I can send her the question, and then if I'm scanning quickly between meetings or, uh, juggling, uh, a few would catch my eye, "Save ten hours a week with this tool, no tech skills needed," or, "From ca- from chaos to clarity, one dashboard to run it all." "Your clients don't need another email, they need this." And sh- this persona actually explains why, even why every one of these headlines will catch their, uh, attention. And I can take the same thing and try it on a completely different persona, like Aiden, and Aiden will give me complete different answer. Aiden will say, "I need one that respects my time and speaks directly to the pain I'm feeling in that moment. Still doing admin during edit days? Here's how to reclaim five hours a week." Other variations that might grade me, "Win better clients without burning out," and so on and so on. So each persona actually answers based on the persona that we got from research, and that single persona represents thousands of potential customers. And so you can try ad headlines, or you can try, uh, a product journey. What would be your f- your best first impression when you get into a new CRM? Um, uh, what would be the feature that will, uh, convince you not to churn a CRM? So you can try it on them. They are twenty-four/seven ready to talk to you on whatever, and I, I really like them. I mean, personalizing those personas has changed the way we work with them.
- CVClaire Vo
I just love this workflow, and to recap it for folks, you took a bunch of, I'm presuming, pretty expensive research-
- MPMichal Peled
Yeah
- CVClaire Vo
... that probably sat in a bunch of PDFs and docs where, you know, we occasionally said, "Head down Hayes, Hayden," but otherwise did not use these personas. You used NotebookLM to create a prompt that embodied the personality of the persona. You put those personas in GPTs, and now you can see that dozens of times your colleagues have gone to them to brainstorm with the persona, which I think-
- MPMichal Peled
Yeah
- CVClaire Vo
... is really interesting, and it's giving me a lot of ideas. So many people go to just plain ChatGPT, and it's like: Give me five headlines for an ad campaign-
- MPMichal Peled
Yeah
- CVClaire Vo
... as opposed to going and sitting with, you know, sitting with your fake persona and saying, "What, what, what ad campaigns would work on you?"
- MPMichal Peled
Yeah.
- CVClaire Vo
So I love this flow. We learned a little bit about the strengths of NotebookLM, uh, GPTs, and
- 41:02 – 48:15
Creating a calendar automation for parking during baseball games
- CVClaire Vo
flipping these, like, sort of personas on their head. Let's go to workflow three, which-
- MPMichal Peled
Yeah
- CVClaire Vo
... I will tell you, I personally, I have a personal connection to.
- MPMichal Peled
[chuckles]
- CVClaire Vo
So people in San Francisco-
- MPMichal Peled
Yeah, we know
- CVClaire Vo
... listen up.
- MPMichal Peled
We are.
- CVClaire Vo
Here's the use case for you. Let's jump to your last use case, and then we can get you out of here.
- MPMichal Peled
Well, this one is actually, yeah, a, a really big pain point.
- CVClaire Vo
And favorite. [chuckles]
- MPMichal Peled
So, yeah. [chuckles] Favorite to solve. Um, well, imagine getting ready in the morning, driving to work, uh, already planning ahead for your busy schedule and morning routine, only to discover that parking in your favorite parking, in your favorite parking lot now costs forty dollars an hour instead of the usual fifty dollar for the entire day that you paid so far.
- CVClaire Vo
[chuckles]
- MPMichal Peled
So this can ruin your entire day, for sure. Uh, so the thing is, HoneyBook's office is, uh, right next to, uh, Oracle Park-
- CVClaire Vo
Uh-huh
- MPMichal Peled
... where the San Francisco Giants play, and on game days, especially those taking place in the morning or afternoon, uh, parking rates spike from fifty dollar a day to a forty-plus dollar per hour. Our team was constantly, uh, getting caught off guard, showing up to expensive parking or scrambling to find remote, cheaper alternatives. We needed a way to know in advance when to take public, uh, public transit instead of driving to work. And so the solution was, I was thinking: Okay, I think let's share a calendar, like a joint calendar, that just show you on which days, uh, parking lot prices are likely to, to, to surge. I needed two things for that. I needed to figure out when games are taking place in the ballpark-... and I needed to create a calendar file. I had no idea how to do. Calendar file is a ICS file.
- CVClaire Vo
Mm-hmm.
- MPMichal Peled
This is the type. I have no idea how to, how to create one. Okay, whatever, uh, let's go to ChatGPT.
- CVClaire Vo
So while you're getting this up, I am just smiling and laughing, because my LaunchDarkly office was right behind Oracle Park, and I got ver- ... I found a $20 a day parking, and I still have... Like, I texted my friend the day I had to pay, like, $100-
- MPMichal Peled
Yeah
- CVClaire Vo
... to park. [laughs] And I was already down there, ready for a meeting. And so San Francisco downtown is, we're coming back, people, but don't forget that the summer, [laughs] the summer baseball season... And sometimes they have two games a day. They have double headers.
- MPMichal Peled
Yeah.
- CVClaire Vo
It's crazy. [laughs]
- MPMichal Peled
Yeah, yeah, yeah. That is correct. And then you have, yeah, walk, walk over there.
- CVClaire Vo
Yeah.
- MPMichal Peled
Just don't use your car. [laughs]
- CVClaire Vo
Just don't go. Just don't go. [laughs]
- MPMichal Peled
Just don't go. If you can avoid it, avoid it. Um, yeah, so I was like, "Let's try ChatGPT." I mean, this should be a simple one, hopefully. So I tried a naive one. Okay, as you can see, this prompt doesn't tell the chat you are this or that. I was like, "I have a simple question: Find, uh, all home games that take place in Oracle Park in San Francisco during the next six month." I used six month because I knew it's, uh, the end of the season coming soon, so you can ask for the next year, whatever. "Filter out only the games that start anywhere between morning to 2:00 PM," because if, uh, games are taking place in the evening, when we arrive in, in the morning, the, the prices are still the usual same one. Uh, "So using these dates, create an ICS, ICS file for Google Calendar," that's the calendar that we're using at work, "uh, that will show these dates as an all-day event." I wanted, I wanted just to see very clearly, um, potential dates, days in my week where I rather avoid driving, uh, to the office. And a key point was availability free, otherwise, this all-day event will just, uh, block my entire day, show me as busy, just because the Giants are playing. Uh, also, "The event description should contain the game details and time." I wanted to add that so I can verify that the game is c- uh, is the one that I'm thinking about, that it's actually one that is taking place there. I, I like to add those extra verification point, validation point, just to make sure that, that, uh, we know what we're talking about. Um, and then I also added an instruction, "Other than just, um, output the ICS file that I need, the calendar file, I want a textual list of all the dates, times, and events included in the created calendar." Now, basically, if it was a human, they may have been a little, you know, offended by me. "Why don't you trust me?" [laughs] But, but ChatGPT doesn't care. [laughs] So it, it thought for 36 seconds, provided me with a file, and also with a list of all the remaining games, because the season is about to end, all the remaining games that are taking place in Oracle Park with their, uh, dates and times. And so I know all of these are included in this file. I just took the file, I, um, installed it or added it to my personal calendar or work calendar. I also shared it with all of my, uh, team members, and then you upload it, and then you can see, for example, that, uh, on September 10th, there is a game. Uh, Arizona Diamondbacks are, are playing, uh, the Giants at, first pitch is, uh, 12:35. So, [laughs] better avoid driving to work that day. [laughs]
- CVClaire Vo
I, I love this so much, um, because, again, I have hit this problem so many times, and you don't want a calendar that has the game in the middle of your work day.
- MPMichal Peled
Yeah.
- 48:15 – 58:45
Lightning round and final thoughts
- CVClaire Vo
This is a great little workflow, I think like a very good little helper personal workflow that helped you and also your team, so thanks for showing me.
- MPMichal Peled
Yeah.
- CVClaire Vo
So just, again, to recap your use cases, first one we did, oh, agents for recruiting. Loved it.
- MPMichal Peled
Yeah.
- CVClaire Vo
So straightforward. I'm gonna use that right away. Two, generated persona GPTs, and three, make your daily life a little bit easier by giving you ambient information that can help with your commute. So we're gonna wrap our episode with a couple lightning round questions and then get you out of here.
- MPMichal Peled
Sure.
- CVClaire Vo
The first one I have to say is, you are the little helper.
- MPMichal Peled
Wow.
- CVClaire Vo
You seem to be all over HoneyBook just helping recruiting, helping the product team, helping the whole team.... you know, tell us a little bit about-
- MPMichal Peled
Yeah
- CVClaire Vo
- your role and what you think this role will look like. Do people need a dedicated person or a dedicated team towards these automations? What do you think the future of this inside companies is gonna look like?
- MPMichal Peled
Okay, for sure. Well, I, I love no- nothing more than talking about [chuckles] myself.
- CVClaire Vo
[laughs]
- MPMichal Peled
So [chuckles] my title is, uh, Technical Operations Engineer, uh, but it encapsulate a lot of other things. So I do, uh, I research and in- i- integrate paid tools, but a lot of the times you don't find the exact paid tool that you want, so I build them. I build internal tools and processes. I'm using no-code solutions, automations, and also coded solutions. It can be an AI-powered Slack bot, uh, it can be an internal application, it could be integrations between, um, two different applications that don't speak with one another, so I come in the middle, and I connect them. Um, it's not just doing things for others, uh, it's also teaching and enabling others to do for themselves. I'm a great believer in enabling, uh, so I do company-wide presentations, I do personal advisory, training classes, documentation. Actually, um, uh, as HoneyBook is, um, is a platform for, uh, small businesses, okay? Um, I see each team and department within HoneyBook as a small business of its own. They provide services. They collaborate with other teams, um, other businesses. Uh, they have their own goals, they have their own expertise, passion for different things, and they all want to spend less time on manual, thoughtless, repetitive tasks and more time doing what they love. So this is where I'm coming for. This is what I'm trying to do. This is what I'm trying to provide, to take the, to take the, the, the friction away and leave you to do what you love.
- CVClaire Vo
One thing I wanna call out for folks is, I've been in tech a long time, and unfortunately, basically up until the last couple of years, I feel like internal tools teams were very starved for resources and occasionally starved for respect. It was like: Oh, you got the product teams, and they're customer-facing, and they build all the cool products, and, like, internal teams are always underfunded, not enough people, blah, blah, blah, blah, blah. And I think now what I love is this is the moment for internal tools teams to shine, to do legitimate, great, high-impact product work. I would recommend anybody who really wants to lean into AI find their way into this kind of role because honestly, a lot of times it's moving faster than you can even get some of these AI experiences into product, which have a lot of, like, customer impact and legal implications, and blah, blah, blah. But if it's, if it's all internal tooling, you can kind of let it rip. And so I just wanna, like, shout out to all the [chuckles] internal tools teams out there that I know today have not got the love and respect that they deserve. This is your moment. Um, you can have really high impact and do some pretty great work and honestly do a lot to differentiate your career in, in this moment by taking advantage of the fact that you can build these tools. So I think you are a great model, and I'm excited to see you do it.
- MPMichal Peled
Yeah.
- CVClaire Vo
Okay, last question. You're a very good prompter. In fact, you create prompters to create prompts. Um-
- MPMichal Peled
One prompter.
- CVClaire Vo
But when AI is not replying to you the way you want, when it is frustrating, when the agent gets distracted by the notifications in its inbox-
- MPMichal Peled
Yeah
- CVClaire Vo
... what is your prompting technique? Are you all caps?
- MPMichal Peled
[laughs]
- CVClaire Vo
Do you yell? You do.
- MPMichal Peled
Wow! Well, I, I love using all caps, and no one-
- CVClaire Vo
[chuckles]
- MPMichal Peled
-can persuade me otherwise. I mean, there will be people saying, "It's just a robot. It doesn't care if it's caps or not." [chuckles] But I'm like: No, it takes me a lot more seriously when it's all caps.
- CVClaire Vo
[chuckles]
- MPMichal Peled
Um, but, but I will say, um, my go-to technique would be to take my current prompt and then tackle the ChatGPT with my prompt, asking it to make it better. And how do I do that? It's not just, "This is not working, make it better." Even as a person, I, I, um, would... I have no idea what you want from me. Um, so I, I'm just going with, "This is the prompt I'm using. This is, this is what's wrong with the output. Like I outline, the output is inconsistent, uh, contains too many hallucinations, invent things that are not there." That's the second part. And then I, uh, it's very important for me that the prompt will, one, two, three, four... I, I list the things that not just what, what is wrong, but what, how I want it to be right. And then I also add, um, I also add it, I give it permission, "Take away everything that doesn't work well." Yeah, "You can delete things from my prompt, you can rewrite things that don't work well, and you can add things that you feel are- will do a better job." And I feel like giving permission to change, delete, remove, whatever, uh, provides a better output because otherwise, and ChatGPTs tend to be pleasers, they will- may try to, uh, use your prompt and not move a lot out of it. Like, "This is yours. This is so great, I'm not going to change it." "No, I allow you to change it. I allow you to r- rewrite it completely." And I tried it several times.... on several prompts, not just my own. People are coming with me with, "Why does my custom ChatGPT act so badly?" Um, and I'm like, "Let's, let's take your prompt and rewrite it using ChatGPT." So I go with that template, and, and, like, first try, it's amazing. First try, you get a l- a prompt that is much, much, much better, and usually it's on- it only takes that one iteration for it to work exactly as you wanted it to. Um, so that's my tip. [chuckles]
- CVClaire Vo
I, I love it. It's a very professional tip, um, that I will use in my moments where instead I just wanna write, "No," in all caps.
- MPMichal Peled
[chuckles]
Episode duration: 58:45
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Transcript of episode -lMItuklFco
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome