Episode Transcript
[00:00:00] Okay, I have a very important discussion to have with you guys today. And it's the fact that prompting is programming. So we are now all computer software engineers. In other words, using AI tools like ChatGPT, Microsoft Copilot, Galileo, etc, you, through your prompts, through your questions, through your inquiries, can program or tell or teach this tool what you want it to do. It's very analogous to the 25 year old or 30 year old birth of the spreadsheet. For those of you that are old, as old as me. When Multiplan, then Lotus, then Excel were launched and we realized that we could build a model and we could save it, and then we could build macros and we could make the macros do different things so that instead of having to recreate and regenerate calculations over and over again, we, we could store them and reuse them. And this to me is one of the most miraculous changes in the IT landscape we've seen since the birth of visual programming. In some sense, all of these integrated development environments, they're often called ides like cursor and lovable and others that are coming out are simply layers on top of your own ability to teach and inform and program your own AI systems to do what you want them to do. And what this essentially comes down to isn't what I would call prompt engineering, because I think engineering is intimidating word, it's learning how to talk, to explain, describe and direct these AI systems to do what you want. Because what's really happening when you talk to the AI platform, the LLM or the lrm, is it's interpreting your English language or whatever language you speak and turning it into commands and essentially a software program of some kind. I don't know what the intermediate language may be. To execute a series of queries and analyses using the L LLM or the lrm. Let me give you just a whole series of examples here because we're going through this with Galileo right now. When you buy Galileo, you get 300, I think roughly 300, maybe more, built in prompts, which are basically mini applications for hr. And each one of those prompts was developed and tested against the data, managed the data that we have in Galileo. And obviously it'll work against your data too. When you put your own HR data in there and they are visible to you. So you can program, you can tweak them, or edit them or add to them, or compliment them. And so when you, when you get a new AI system, it's not just an off the shelf application, it's a programming environment, it's a development tool, it's a, an automation tool, a robot that you can use to do whatever it is that you do for a living. Now, you know, most of us are not engineers. I happen to be one. And so, you know, a lot of you I know don't want to spend a lot of time creating stuff. They want to, you want to use it. But let me give you just sort of a sort of workflow of why this is so useful. So let's just go through the life cycle of an employee. In the human capital management domain, the first thing that happens is you hire somebody. In order to hire someone, you have to create a job description and you have to create a job title, a job level, sort of a pay band and responsibilities. And the way you do that is, is by interviewing managers or copying a job description you had in the past that you've been using. So normally when you don't have AI, that's a manual process and you might tweak it and edit it and make sure that it's not biased and it doesn't have anything illegal in it. So there's some tools to help you with that. I would suggest that what a lot of you can do is develop a simple prompt. We have, by the way, a bunch of these in Galileo that would say something like, Please interview the following five managers. Insert email address or phone number here. Consolidate those interviews and create a list of the responsibilities, skills needed and most important tasks in this job based on the five people that you just interviewed. Write it into the form of a job description. Go out and query Galileo for the job titles that most closely associate with this job description. Pick the title that is the most closely associated with this job description in the external market. Go out into Galileo and look at the average pay levels for that job in the city in which you're hiring, whatever that city may be based on tenure. By the way, all that data is in Galileo and bring this back and write me a job description. Now once you've done that, take our corporate values and our corporate principles, which are located in the document that I just uploaded, and apply them to this job description so that the applicant understands what the company is like. The culture, the expectations and the work environment to that job description are standard practices for flexibility over time. Scheduling, benefits. Use the benefits for the job level of this job. By the way, I left out level earlier. Let's we could include that in the pre prompt and generate this job description and then send it to the five people that you interviewed in the beginning, get their input, collect their input, and iterate through the job description again and then send it to me, the HR business partner, to prepare for posting. Now that took me two minutes to say into the podcast, you could say it into the AI verbally, the AI would transcribe it and you could run it. And it might take a few tweaks to get it to work perfectly, but more or less, you've just automated yourself. Three weeks of work, two weeks of work, a huge amount of work, and you did it yourself without going to it. And if you did that once and you're a recruiter and you got it to work well, or maybe you're sort of a techie that works in recruiting and you circulated around your team and they used it within a week or two, you could make it exceptionally perfect because it could act your email system, it could access your corporate values and so forth. So that's number one. Number two, after you've assessed these candidates, or you get a list of candidates from whatever system you're using and you get a whole bunch of resumes back, and maybe each person that was selected through that resume screening Process gets a 10 minute automatic interview from one of the automated AI interviewing systems, including Eightfold Maki People, Paradox or others. Take the audio files of those recordings, put them into a folder, ask say to the system, go through those recordings and now please compare the capabilities and culture fit and availability of these candidates against our culture document against our job description and give me a rating of each candidate on a one to five rating on how well they fit against those characteristics, one being no fit, five being an almost perfect fit. And make sure that you evaluate the candidates in a form that only 10% get a 5, 20% get a 4, remainder get a 3, 2 or 1. It will do that for you and you'll get a first cut of who the best candidates are. So you've now done some amount of screening, some amount of assessment, some amount of filtering. The next step is probably to talk to the people on the phone. And there are conversational interviewing tools for that, but you probably want to do that yourself to get to know them. And once you get to know them as a recruiter or as a hiring manager, you might want to have other people interview them and you could have a tool that does that. So let's suppose you're ready to make this person an offer. You've picked the right person and now you say, I want to negotiate a start date, a starting salary, benefits and working conditions for this person and you take these standard job characteristics like that that you've used for the last 10 people who had this job and you put that together and you say, please write this into an offer letter for this individual. Make sure in the offer letter you also discuss this person's particular interests and some of the areas that they may be strong and some of the areas they may be weak and make it a positive letter in a form that would make them want to come work for us. And sure enough, I'm sure it'll wr a great letter for you. Maybe perfect, maybe not quite perfect. Go back, fix that. Okay, now you've got an offer letter. The employee decides to join your company. Hooray. You're all excited. You want an onboarding program. You have a standard onboarding program already of, you know, 17 things. Getting a new computer, setting up your password, finding a badge, getting your desk and so forth. And then you say to the hiring manager, so and so is starting on this date, would you please give me your, your feedback on some of the other things you'd like to do for onboarding, such as what is the information you'd like them to read in advance? What is some of the training you'd like them to. Who are some of the people you'd like them to talk to? You take that information from that conversation. You ask the AI to put it into an onboarding plan with step by step by step, you store that as another little program. Your onboarding program you generate. You now have an onboarding program for this person. The person starts the job, maybe you have an onboarding tool, you know, in one of your HCM systems, maybe you don't. And you know, a month in, the manager is expected to give them the first month or the first 60 day review. And so the manager sits down with them and either individually or we give them a chance checklist and they walk through and discuss what are some of the things this person's done well? What are some of the things where they're above expert expectations, what are some of the things where they still need to spend more time learning? Who are some of the people they should talk to? What are some of the opportunities they have for improvement? We get feedback from the employee on what they like about the job, what they don't like about the job, what they want, help with what they want to do, more of what they want to do, less of that. All gets recorded, stored digitally. And now you have a development plan for this person. And you ask the AI in the third or fourth little program you write here, please write a development plan for this person based on that conversation. That is another little GPT or a little task we call it in Galileo, that you could create. Give that to the manager, give that to the employee. You now have a plan and you can see how this goes. The next step is creating a one year performance review where we take the framework of the company's performance, process, the employee's goals, and we create a prompt or a program to generate a standard performance review document in a positive way, talking about the company's culture as well as the employee's results and activities. Then the person's ready for leadership. We, we go back and we look at all the information we have about this person through all the conversations we've had and all of the documents we've created. We assess them against our leadership model, our corporate leadership model, or the Hydric leadership model, or the UCF SHL leadership model, most of which are in or the irresistible leadership model, all of which are in Galileo. We give the person a readiness fit whether they're ready for leadership as well as a development plan for their leadership first leadership responsibility, on and on and on. I mean, this is programming.
[00:10:44] It's not just asking questions and getting queries like in Google. This is a programming tool. Now, the reason I'm explaining this in such gory details is that when you look at what the software engineering companies are doing that you're spending money on, this is what they're building. They're building under the covers, complex, fairly well tested, but not perfect prompts designed to use the data that they have and the data that you have in the structure that it's been created to answer these questions. And a lot of you are going to do this yourselves. And I think one of the most interesting elements of the AI transformation in business is not the fact that we're going to buy a bunch of tools and they're going to be fantastic and we're going to turn them on and they're going to suddenly replace all sorts of work for, but that we're going to be able to customize them and configure them and train them and tweak them and learn from them and teach them what we want and they'll get better and better and better. And even though today only 6% of the workforce are software engineers, it's somewhere around 6 or 7%. It's pretty small. I would imagine that we're going to have more like 10, 15, 20% or more of workers. Many of us who use spreadsheets will probably gravitate to this that are going to be building things with these AI systems. And what that means is a couple things. First of all, it means that the vendors who sell us these platforms need to think about them as development tools, not only off the shelf applications. We have to give customers or users the ability and the tools to tweak the solutions. Give you an example, I was using Lovable, which is a IDE for websites, and I asked it to build a very simple landing page, which is a webpage for me, just as a test for a certain research report we had. And I gave it all sorts of information about this page and, you know, what it needed to look like, what it needed to say, and where it should get information from it. And it sat there for 15 minutes creating this thing. And when it created it, it was pretty good. It wasn't perfect, it was pretty good. I went back and asked it a few more things, asked a few more things, and eventually it got to a point where I liked it. And what I was hoping the thing would do is give me a way to save the complex series of prompts I had given it so that I could reuse that. It didn't seem to be able to do that, at least not clearly to me. Maybe that's something I just missed. And to go back and see what I had said and tweak it to make it better and better. And those are the kinds of things that I think those of you who are vendors, and I suppose in some sense we are too, are going to want to give to our customers. For you, as an HR person, as a business person, as an IT person, you are going to have to think about these tools as development tools. So experimenting with them, playing with them, tweaking them, building things is going to be one of your core competencies, not just dreaming up questions and randomly shooting them at the AI to see if they give you the right answer. But you're going to have to think about what data is in there. How is it going to answer this question? What do I need to add to it? Just like you would in an Excel spreadsheet. You know, any of you that's built, that have built complex Excel spreadsheets, you know that, you know, you might spend a couple hours building it and all of a sudden it's got an error because somewhere off in the corner it's accessed a piece of data that was either typed in incorrectly or missing and it's giving you the wrong answer. And those are very, very, very Very powerful tools. I mean, I can't imagine the amount of manual labor that was eliminated with Excel. I mean, Excel was probably the most fantastic automation tool ever create.
[00:14:05] So this is sort of a two or three orders of magnitude more than Excel was. And we're going to be developing these things now. The third thing I think that comes out of this is given this highly unstructured potential we have for these AI systems, how do we implement this in a company? Because if we ask each employee to figure out on their own how to use these tools, some percentage of them will, but a lot of them won't, and they won't even really know where to start.
[00:14:31] So the answer that's becoming more and more clear, especially today, after there are a few more layoffs from Microsoft, a few others, is for so those of us that are sort of systems thinkers and engineers in our hearts, and you know, you know who you are, there's people that like doing this. We need to sit down with our teams and offer to build these reusable, repeatable programs or prompts, or agents, whatever we want to call them, for others and take it upon ourselves to share them, just like the person that built the budgeting spreadsheet probably did from the finance department many years ago. And because this is an English language and we don't have to really go to college or graduate school to learn how to do this, this is going to encourage all of us to have systemic or system thinking, to think about the data that we have access to and where we're going to get it, and to learn how to think programmatically about what do we want these things to do. In some sense, building a prompt is very similar to clearly, clearly, clearly defining a problem, which is not always easy. Sometimes, you know, you might be in a meeting and somebody says, why don't we do such and such, and you think, okay, I'm going to go work on that. But. But you really need to sort of brainstorm within your own mind or within a few people. How are we going to solve this before you just kind of throw something into the AI? And the nice thing about AI is it's iterative, so you can develop this over time. The IATA airline industry skills model I talked about last week, which we're going to spend more time on later, is an example of this. The 300 or so prompts that come out of the box in Galileo are all designed by our team with a lot of experience. We know how to do this. We've been doing it for a long time. We've seen a lot of client implementations. We know all sorts of data sources that are in there. And you also have to know what data your AI is connected to. In the case of Galileo, we tell you very explicitly it's not connected to the Internet. So you're not going to get any random chaos in there. But in your own company, if you connect it up to one of your CRM databases or your Monday.com database or your finance system or some, you know, big Excel spreadsheet, you as a programmer, as a user are going to need to know what state is in it. Anyway, I'm sure a lot of interesting books will be written on this. There's probably dummies books right now on how to do this. But I wanted to sort of inspire you as we go into the holiday weekend here, think about the power you now have as a super worker, being able to use these tools in your own unique human value way to solve problems for yourself and solve problems for others in your organization. Big change in the market, a big set of empowerment here and a massive unlock of productivity growth and innovation in our companies. That's it for now. See you guys next week. Bye.