Episode 1 - The Rise of AI Personas

00;00;15;01 - 00;00;37;17
Jie Ren
Hi everyone. Welcome to my podcast: When Tech Meets Ed. I'm Professor Jie Ren. I'm very happy to have my PhD advisor, Dr. Jeff Nickerson, to join us to discuss a very important and interesting topic: AI Persona. I know Jeff recently went to Busan for a conference. So, Jeff tell us about the Busan trip.

00;00;37;20 - 00;01;12;03
Jeff Nickerson
Yeah, so I went to a conference in Busan, called the User Interface Software and Technology Conference. It’s an ACM conference, (Jie Ren: Amazing) mainly computer scientists who are building things. And I was part of the organization for a workshop on Longitudinal Research and AI. (Jie Ren: How cool). Tao Long at Columbia ran this workshop, and it was very interesting, because a lot of the conversation was, does it even make sense to do Longitudinal Research in AI because things are changing so rapidly?

00;01;12;06 - 00;01;40;08
Jeff Nickerson
One of the points that many people at the workshop made is that it is changing very rapidly, and yet, since it's going to be maybe the future of the way we work. It’s probably quite important to not just have 5- or 10-Minute interactions with AI and write papers about it, but to maybe have interactions with AI over days, weeks and months, and start to understand how, how it changes the way we interact, the way we work.

00;01;40;11 - 00;02;01;07
Jie Ren
That is amazing. I know we are going to talk about AI persona, so, but given what you have described, right, I want to take a little bit shift in terms of, you know, like, AI models are being advanced so quickly, etc., etc. So, what do you detect? Issues or, you know, good things related to one AI model. The good things will stay there,

00;02;01;07 - 00;02;14;10
Jie Ren
but the problems or something, they will disappear. So that definitely will change. You know, how we use, for example, AI persona and an agent maybe, like, you can, like, talk about it briefly, and then we can just move to, you know, your background, etc.

00;02;14;14 - 00;02;37;29
Jeff Nickerson
Yeah, so, so, I think one of the really interesting things that the technologies are changing very rapidly, and one of the interesting things, relevant to what we're going to talk about in terms of in terms of work, is that when these AI models came out, I remember a long time ago when Amazon allowed you to talk to an Agent

00;02;38;05 - 00;02;39;07
Jeff Nickerson
Yeah. That.

00;02;39;10 - 00;02;40;20
Jie Ren
Yeah Alexa

00;02;40;23 - 00;03;10;26
Jeff Nickerson
Right, exactly and those conversations would only last about, like, 20 seconds before they would go totally off track, right? And now we can have conversation with these agents that go back and forth for several minutes, right? (Jie Ren: Yes) The newest thing are agents that will go away for, say, 15 or 20 minutes, right? (Jie Ren: Mmmm ok) So, for example, the coding agents I can ask, and I've been doing this, I've been doing this recently where I've been talking to an agent

00;03;10;27 - 00;03;22;08
Jeff Nickerson
(Jie Ren: Yeah) And I've been asked it to build a software system, and it will ask me a few questions, it will go away for 15 or 20 minutes, and it’ll build me several 1000 lines of code, right?

00;03;22;09 - 00;03;22;24
Jie Ren
Amazing.

00;03;22;25 - 00;03;45;04
Jeff Nickerson
So, this is like, really different (Jie Ren: Yeah). Now yesterday, open AI announced that they have a new coding agent, which they claim can go for an entire day without intervention. (Jie Ren: Nice). Now I have not tried it yet, and (Jie Ren: Okay) the online conversation is a little bit skeptical that it could actually go an entire day, but I think that's the direction

00;03;45;06 - 00;04;09;03
Jeff Nickerson
(Jie Ren: Yeah) Is it the delegation? Basically, what's happening is it's like delegation with humans (Jie Ren: Yeah) is you don't want to delegate, certainly, you don't want to do 32nd Delegation. And even five-minute delegation is too short. And even an hour delegation for a human being is too short, you know, really, you want to get up to a day or ideally, like a week.

00;04;09;05 - 00;04;23;12
Jeff Nickerson
And so, I think that's what the I think that's what we're headed toward, are agents where you can kind of say, hey, go off and investigate this and give me something a week later, rather than give me something in like 20 minutes or so.

00;04;23;14 - 00;04;38;10
Jie Ren
Nice, so given like, AI agents be more and more productive, right? So definitely, they can be applied to many fields more productively. Right? We're going to talk about that in a bit. So now I want you to introduce yourself. You have so many titles, I don't know where to start.

00;04;38;12 - 00;05;08;06
Jeff Nickerson
So, right now, and I'm an emeritus professor at Stevens Institute of Technology in business school, and I did many things before that. I think pertinent to our conversation is my Undergraduate and first Graduate Degree were in Visual Design and Graphic Design. (Jie Ren: Yes, design) And so I have a design background. I started working with a design background, and later I went back and I got a PhD in computer science,

00;05;08;09 - 00;05;14;15
Jeff Nickerson
and a lot of my work has been trying to figure out how to integrate these multiple interests that I have

00;05;14;17 - 00;05;29;03
Jie Ren
Nice, and now, like under your supervision as a PhD student, I started to get into this field that is called collective intelligence and creativity. So why don't you start with that? Like, how did you start your journey on this topic?

00;05;29;05 - 00;05;52;27
Jeff Nickerson
Yeah, so I got interested in the way I got to collective creativity was the inception that really came from a conference workshop on information and networks, which was held at NYU. And it was a very interesting workshop, essentially, with about 100 people every year. And it had a mix of people, had Business School people,

00;05;52;29 - 00;06;22;20
Jeff Nickerson
but it also had Physicists and Computer Scientists and Sociologists who were mixing together everything they knew about networks, right? So, sociologists been working with social networks for a long period of time. But also, Physicists became interested in social networks because for a Physicist, you know, you have atoms, and atoms are connected to each other, and you and then you get chemistry and like that, and so and so. For physicists, this is also a very natural thing to study,

00;06;22;22 - 00;06;55;29
Jeff Nickerson
and for Economists, it was also very natural to study because you have marketplaces, which essentially can be modeled as networks of individuals. So that became, became a place where I realized that I could do something that was pertinent to Information Systems that involve my interest in networks. But then I had another puzzle. Is that I wasn't really interested in studying, as an example, the impacts on finance. I was much more interested in, can the networks be used to create new things?

00;06;56;01 - 00;07;22;12
Jeff Nickerson
And so, I found that there is an area of research called collective intelligence and within collective intelligence, there was kind of a burgeoning area called collective creativity, which said, can we, can we connect in this, in the initial stages, you know, sets of people together that were that were very large, and have them create things

00;07;22;14 - 00;07;37;20
Jeff Nickerson
and this was kind of the inceptions of crowdsourcing. Can you crowdsource certain aspects of things? So instead of having, you know, one person creating something, or five people creating something, you could have hundreds or even thousands of people creating things.

00;07;37;23 - 00;07;58;29
Jie Ren
I totally agree with you, because, like with crowd, right? It’s not like one person, as you said, or a small group of people. It's like 1000s of people, and how do you organize them for better outcome here, creativity, right? So now you can use the network, right, and also distribution system to organize them, like, in different ways, to see, like, which way could like better generate something,

00;07;58;29 - 00;08;09;19
Jie Ren
right? So, can you also, like, talk about, like, how you focus on AI persona? (Jeff: Yeah) You see the natural kind of transitioning, (Jeff: Yeah there) but I want you to [inaudible]. 

00;08;09;20 - 00;08;35;21
Jeff Nickerson
There is a there is a transition. Because, like, as we got large language models, it started to become apparent that a lot of the things that we asked Crowd Workers to do, you could also ask a Large Language Model to do. And in fact, one of the other conferences that I go to, and I'm actually on the Steering Committee of Collective Intelligence, took this up recently at a conference in Delft.

00;08;35;24 - 00;08;52;28
Jeff Nickerson
One of the things that was there were, there are two things that were amazing about the conference. One was that there were a lot of people who had been doing Crowd Work, started to, like, start calibrate and go back and take all the old Crowd Working papers and see if you could get Large Language Models to replicate that work

00;08;53;01 - 00;08;53;17
Jeff Nickerson
and the [inaudible.

00;08;53;20 - 00;08;57;18
Jie Ren
Is it better than the crowd work, or, like, worse. What’s the comparison?

00;08;57;18 - 00;09;14;29
Jeff Nickerson
Well? So, the main thing is, could you get it to even do it. So that was, like, really the thing? And if you could get it to do it, like, how quickly could you get it to do it? Like, is it more or less expensive to hire AI Agents to do it? And the, you know, the result at the time, so this is several years ago,

00;09;14;29 - 00;09;33;08
Jeff Nickerson
the result at the time was that not all Crowd Work. It wasn’t easy to get all Crowd Work replicated, but some of it you could replicate, and you could do it quite quickly and it meant that you could move more quickly, because, as you know, having done Crowd Work, it can take like, you know, you have, you have to ask people to do it,

00;09;33;08 - 00;09;36;11
Jeff Nickerson
and people like, not always are available, and they're

00;09;36;11 - 00;09;37;05
Jie Ren
Reluctant.

00;09;37;05 - 00;09;59;22
Jeff Nickerson
And you want them to come back later and do things again, and they don't want to come back and like it's hard to manage, and with Crowd Work, like AI agent says, Crowd Work is much easier to manage. So, there’s a sense that maybe this is possible. Then the other thing that happened, which is the most important thing for my work in Persona, is there, there is a workshop at the conference called Crowd Camp.

00;09;59;22 - 00;10;23;17
Jeff Nickerson
Crowd Camp has occurred at a number of different conferences. And at Crowd Camp, a set of us got together and said, could we actually create a, could we use a Large Language Model and create Persona, you know, have the Large Language Model, create Persona that would have conversations, that would be generative in some way to either solve problems or generate things.

00;10;23;20 - 00;10;41;26
Jeff Nickerson
And we tried this out as a workshop. We just tried out this idea, and it worked, right? So, there was we, we got, like, really interesting conversations out of it, and we also got productive work. And that was the moment in which myself and other people involved in it said, there's really something here.

00;10;41;28 - 00;11;04;01
Jeff Nickerson
And we started saying, Well, maybe these personas are, are kind of the equivalent of the crowd worker from before, but they have really interesting aspects to them. So, so as an example, what was really interesting is, there were four of us at the time, and the four of us created four agents.

00;11;04;04 - 00;11;24;14
Jeff Nickerson
(Jie Ren: Okay.) And we asked them to do design, and we monitored their conversation. And then we ourselves, as we're monitoring, are talking among each other. So, it's kind of like we have a shadow team. (Jie Ren: Yes) That's doing design, and we ourselves are talking among ourselves. And then we're trying to figure out, kind of, as the managers, what do we ask them to do next?

00;11;24;16 - 00;11;31;05
Jeff Nickerson
And so, this, in some sense, becomes a paradigm for the way one can use AI persona to do things.

00;11;31;07 - 00;11;57;20
Jie Ren
Yeah, that's amazing. That's very much like AI agents, you know, more related to the job functions, etc., etc. Now you can add certain personalities and then having the conversations. Maybe they could be working individually, or they could work in a team, right as you guys are having conversation a human, human, like supervisor, having conversations, and then guiding this, like AI agents with, you know, different personas working together.

00;11;57;20 - 00;12;26;05
Jie Ren
I think that's really cool. And then I also want to talk about, so far, we have talked so much about the academic kind of development and discussion on like, AI persona, etc., etc. I want also to talk about the commercial implications, the business implications, the use of AI personas in the business context. So, for example, right? We could see, you know, AI personas are becoming online influencers, right? Like affecting a lot of, for example, younger people on social media.

00;12;26;11 - 00;12;47;01
Jie Ren
And then we also see the commercialization of AI Personas in terms of helping me or helping anyone to learn the foreign language and practice with these users. Or it could be like AI personas, you know very much, providing online companionship for someone who is having a bad day. So, could you please, like, talk about, elaborate on this business,

00;12;47;03 - 00;12;53;01
Jie Ren
you know applications of Online Personas sorry, ummm AI Personas?

00;12;53;03 - 00;13;15;01
Jeff Nickerson
Yeah. So, I think all the things you mentioned are, are things that are developing right now. And I think they all, I mean, we're the early stages on all of them. One of the things that, that, one of the things I wanted to raise, I went to a, one of the things I like doing is I like doing Hackathons around AI

00;13;15;09 - 00;13;35;26
Jeff Nickerson
and I went to a, I went to a Hackathon in New York. It was part of Tech Week in New York? Well, no, actually, it was, I went to a Hackathon to Tech Week, and then I went to kind of some add-on Hackathon spun out of that. And one of the Hackathons was run by a VC firm.

00;13;35;28 - 00;13;58;22
Jeff Nickerson
And the prompt, whenever you go to a Hackathon, they give you a prompt, and you don't really, you know, you don't really know what to expect until you get the prompt. And the prompt had to do with Persona (Jie Ren: Nice) and it’s like, oh, interesting (Jie Ren: Yes). And it became, it became evident that, you know, that there is been a lot of work in Persona, in the startup community.

00;13;58;28 - 00;14;19;18
Jeff Nickerson
(Jie Ren: Okay) And in fact, it turned out that the VC, what the VCs were really interested in, was that one of the big problems with a company starting is, how do you get the first few customers right? (Jie Ren: Yes) Like, how do you know, because that once you have the first few customers, then you can often, like. (Jie Ren: You can scale up)

00;14;19;18 - 00;14;20;24
Jie Ren
You can scale up right.

00;14;20;27 - 00;14;29;21
Jeff Nickerson
(Jie Ren: Like you hope). You hope, right? And how do you, how do you, kind of, like, get an intuition that you won't be able to just get your friends to use it? (Jie Ren: That’s true,) 

00;14;29;22 - 00;14;31;18
Jie Ren
always the starting point, right?

00;14;31;18 - 00;14;56;09
Jeff Nickerson
Right, there’s often the starting point, but the VCs are much more interested in, they discount that much more interested in, are there? Are there, like, real people out there? Are there Chief Marketing Officers or Chief Technology Officers, or whatever that are going to buy your product, or at least are interested in your product, and so there now is, like a whole infrastructure around kind of, what I will call Customer Personae.

00;14;56;12 - 00;15;14;14
Jeff Nickerson
And you can think about it, you know, you think about, well, you know, a quite obvious thing to do is, if you were thinking about like a mass market retail thing is, you could create people with certain demographics and see if they're interested in an Ad you're going to give them. But to me, a little bit more interesting is in the business to business area,

00;15;14;17 - 00;15;23;09
Jeff Nickerson
can you create Personae who are managers inside companies that can purchase technology, and will they be interested in your product?

00;15;23;11 - 00;15;45;26
Jie Ren
Yeah, but it sounds very much like testing the idea, right, testing this prototype or testing this product to see what is the potential market, or is it interesting to like anyone right, as a potential customer or potential investor. But how would we go beyond that to bring in, like, real customers?

00;15;45;29 - 00;16;23;29
Jeff Nickerson
So, here's the kind of like, I think that's the that is the bridge that these companies are trying to cross. But here's kind of an interesting thing that I noticed from conference, is turns out there's already kind of an infrastructure out there of different tools and libraries related to this, and it seems like there's just two things. Is one is, you want to be able to create a set of persona that describes, say, you know, Chief Marketing Officers. And you want to test your message on these Chief Marketing Officers, until you get a message that seems to resonate with the with the Persona.

00;16;24;01 - 00;16;58;18
Jeff Nickerson
Now question is, can you find a real person that matches the Persona? Yes. And so, there are other companies that specialize in that, right? So, in other words, once you have a description of a Persona, they will look through, say, LinkedIn profiles, to try to find a profile that matches. And now, now you have a Sales Lead that might be a lot better than a typical Sales lead, because you've already, you kind of understand what you're looking for. You're looking for, you know, you might be looking for a mid-market company in the Northeast that is close to a distribution center, as an example,

00;16;58;18 - 00;17;14;24
Jeff Nickerson
like these very, very specific things is what, what your analysis has shown is a customer that might actually be interested in your product. And now you got that, you try to find a person that is like that, and then you can send them a very targeted message, saying, this is why your product will fit what they're doing.

00;17;15;00 - 00;17;20;05
Jeff Nickerson
It's a lot better than a blanket, you know, than us a blanket email or a blanket post on LinkedIn.

00;17;20;07 - 00;17;40;07
Jie Ren
Yeah, it's one step further than just plainly, you know, working on this market segmentation, trying to figure out who your customers are, right? And then in this case, you can actually do the same kind of like doing the simulation right using this AI personas, but actually you are matching them to real people and doing this more targeted marketing strategy.

00;17;40;09 - 00;17;59;08
Jeff Nickerson
Yes, and I'm also interested in another aspect of this, which is, so a lot of people at this, at this Hackathon, seem focused on that kind of getting your first customer type of thing, which is, I think, what the VCs are focused on. But I also think there's another way you can use Persona, which is in product design.

00;17;59;15 - 00;18;02;09
Jie Ren
Oh, yes, yes. About creativity, back to creativity.

00;18;02;11 - 00;18;23;10
Jeff Nickerson
So back to creativity is so you can assemble, you know, you might have in a startup, you might have an R&D team, which is like one or two people (Jie Ren: Yeah) Right. And you want to have a larger R&D team, and so you can create Persona and your R&D team to have them develop product. And if you think about it, what you're really looking for in startup is who is Product Market Fit,

00;18;23;12 - 00;18;43;15
Jeff Nickerson
right? So, someone is finding a market for the product you think you have, but it's also modifying the product so it fits an existing market, right? So, this thing goes both ways, yes. And you can imagine, if you have persona, you have customer persona, but you also have R&D persona, that you could set off this kind of back and forth

00;18;43;18 - 00;18;58;07
Jeff Nickerson
to try to try to evolve, you know, you evolve your product and then your market. You know you might evolve a product to fit a particular market, but then you might ask for a search to see if there are other markets now, now that you changed your product, maybe there are other markets that you haven't seen before.

00;18;58;09 - 00;19;18;00
Jeff Nickerson
Maybe it doesn't have to be in the northeast, maybe it doesn't have to be near a Distribution Center. And so, in some sense, you can go back and forth and maybe expand the universe of people that might be interested in your product by like, looking at the directory of the product, and then also looking at the at the kind of adjacencies in the customer profiles.

00;19;18;03 - 00;19;34;29
Jie Ren
Kind of reminds me of this very optimistic view that you know anyone could be leading a company with AI employees. So, like, what is your view on this? It is realistic to do?

00;19;35;01 - 00;20;00;14
Jeff Nickerson
So, I think it is so. One thing I think is realistic to do is to think differently about company formation, because now we have AI, right? I also think that because, because AI may have, I think it's already having effect on the job market. I think that their entry level students that are graduating from our programs are having a harder time getting initial jobs than they used to have.

00;20;00;17 - 00;20;23;07
Jeff Nickerson
And I think part of that is entry level jobs. An entry an entry level student is competing against AI (Jie Ren: Yeah). And in some cases, the companies are saying, we have AI we don't need this entry level employee in the two years’ worth of training (Jie Ren: Yeah). We can use a more experienced employee managing an AI agent rather than managing an intern or an entry level employee,

00;20;23;09 - 00;20;39;01
Jeff Nickerson
right? So, I think in that kind of job market, if I were a student that was sending out hundreds of resumes and not getting any leads, I would form a company, right? (Jie Ren: Okay) So and I, and I think that part of the reason I would do it is I don't need that much capital anymore form a company. (Jie Ren: That's very true). 

00;20;39;01 - 00;21;00;29
Jeff Nickerson
I need a subscription to one or two of these services, and I need a budget for an API, and I need a decent computer, and I need a, like, a space within which to do this. But that might be, that might be all I need to kind of get things going. Now, the other question is, I still think there's skill involved, right?

00;21;01;01 - 00;21;24;24
Jeff Nickerson
(Jie Ren: Yeah). So, in other words, working with the working with these agents, they’re what, what you find is that the more you tell them, the better you get. (Jie Ren: Yeah) And so, I think like thinking about education, I think one of the things we should be, we should be working with the students, is getting them to get better at, like, how do you even, how do you even think about, you know, prompting an agent.

00;21;24;27 - 00;21;51;16
Jeff Nickerson
If you think about persona as an example, what persona are going to be important, if you're thinking about a structure where you're going to build agents to do some of the work inside a company, like, how do you structure that? How do you delegate? What are, you know, what are the objectives? These kinds of things which we talk about, which we do talk about in, you know, in kind of entrepreneurship courses, I think do apply to managing AI,

00;21;51;18 - 00;22;11;27
Jeff Nickerson
but I also think there are particular skills to managing AI, right, like, I think a big difference is, is in a company, if you're delegating to an employee the pace with which the employee comes back to you is relatively slow. You delegate. You ask them to come up with a few product ideas. You expect them to come back, maybe in a few days, to have a conversation.

00;22;12;00 - 00;22;42;19
Jeff Nickerson
In this case, you're asking and getting something back very quickly and so very quickly. It can be overwhelming the amount of reading that you need to do to even absorb and manage the AI, right? And so, and so that's you know, that's kind of like as an experience issue is, how do you get experienced enough to know, how do you give the prompts and how do you manage what's coming back; And when do you realize that maybe you're getting too much stuff back and you need to do some thinking

00;22;42;21 - 00;23;04;04
Jeff Nickerson
or maybe you figure out ways of letting the AI itself, monitor its results, change its results, and redirect its results the same way a human employee would. So, you're not getting back what's called kind of AI slop, but you're getting back things that are like well-considered and going to be much closer to what you can act on as an entrepreneur.

00;23;04;06 - 00;23;30;16
Jie Ren
So if I do a summary, right? So, for example, if anyone wants to be an entrepreneur by the use of AI Agents or Persona, that person first needs to have a Strategic Vision, that's very important. And then at the same time, needs to have technical abilities, right? For example, how to get connected to API’s, etc. And then the other one is very much like the soft skills that we were, you know, like in the traditional context, educational context,

00;23;30;16 - 00;23;56;12
Jie Ren
right? So, like, anyone needs to have soft skills. But in that case, we are talking to human workers while interacting with human clients, human suppliers, etc., etc. And now we need to apply that to these conversations with AI Agents and AI Personas and more, having a design view in terms of, Oh, how I showed you these AI agents to carry these tasks right

00;23;56;12 - 00;24;15;05
Jie Ren
and maybe like, like from, from this supervisor’s point of view, human supervisors point of view, break the big task into like small pieces, kind of like this person is wearing multiple hats, right, doing multiple functionalities. But again, with the AI agents and personas help, that is doable.

00;24;15;07 - 00;24;45;09
Jeff Nickerson
Yes, I think it is, as you said, like managing AI agents is, is a similar skill to managing humans, but there are differences, right? And in particular, one of the things that that I keep on running into with AI agents is that I will, as an example, I will ask it to build a chat bot, you know, to build a chat bot tool that I want to give to somebody else to use,

00;24;45;09 - 00;25;04;26
Jeff Nickerson
right? So, in other words, I want to build a layer on top of a chat bot, on that that is a as a very focused tool, um, and it will build it for me, and then I will test it and as I'm testing it, I will get a suspicion that maybe it's not what I think it is.

00;25;04;29 - 00;25;21;05
Jeff Nickerson
So, I'll ask it, are you, is this actually a chatbot? Are you actually going out like I told you, are you using my API key to go out to anthropic now? Are you really doing that and it will say, yes. I’ll say because some of those replies, it seems like some of these replies seem a little bit canned

00;25;21;07 - 00;25;43;26
Jeff Nickerson
and it said, nope, that's just the way it is. And then I'm like, hold on a second, and we'll go in and look at the source code, and we'll find that it’s built this very elaborate conversation simulator, would say, with, say, 100 of kind of canned replies. And it's taken me a while for it to cycle around and start using some of the same replies,

00;25;43;29 - 00;26;01;20
Jeff Nickerson
and so then I will ask it. I say, Hey, you're actually using a simulator, right? And it goes, yes, I'm using a simulator. And, and if I ask it why, it’ll say, Well, I tried to use the API, but the API key was not accepted so, in order to make further progress and please you, I created a simulator.

00;26;01;23 - 00;26;21;01
Jeff Nickerson
Okay? Now, in some ways that's right is trying to please me, but in other ways it's not right, because I actually want the real I want it to do the real thing. I don't want it to create a simulation, and it doesn't seem to differentiate between the real thing and the simulation in the same way that I would.

00;26;21;04 - 00;26;41;14
Jeff Nickerson
So now in my prompting, I need to say, look, if you run into an obstacle and you want to use a simulation, come back to me and kind of like, ask permission before you do that, because maybe we can find a way of solving the initial problem, of solving the API key problem, so that we can build a real thing rather than a simulator,

00;26;41;14 - 00;27;00;28
Jeff Nickerson
right? So that's very different, with a human, you ask a human to go off and do something like that, they're not going to come back and tell you that they've got it working if they've built a simulator, they'll come back to you and say, look, I built a simulator because I couldn't get the, that was the first thing they would tell you, because they know that that changes the quality of the solution.

00;27;01;01 - 00;27;24;20
Jeff Nickerson
So, in this way, managing AI is different because it has different, like, it's not grounded in the same way we are as humans. And so, it becomes possible to, if you don't recognize it's going to do that, it might be possible to be talking to this thing for like, several days without realizing that you're not actually, you're talking to essentially a simulation, rather than something doing the work that you expect it to do.

00;27;24;22 - 00;27;41;11
Jie Ren
Yeah, so part of that is Prompt Engineering, but it's not only on that level. And then, as anyone you know who’s who is thinking about using AI agents to do the job, that person also needs to be technical, right? If the job is related to technical stuff.

00;27;41;14 - 00;28;11;26
Jeff Nickerson
I think there is some level of technology that is very helpful to have, like as an example, if you, if you’re just talking to, having a conversation with these agents versus building on top of an API, you are limited to what that interface gives you, whereas if you can build an API, you can create, like, alternative interfaces, which, you know, allow you to tailor your solution and differentiate what you're doing from what other people are doing and maybe go deeper in certain ways.

00;28;11;28 - 00;28;31;20
Jeff Nickerson
So, I think, like, being able to master the API calls is kind of important. I will say that there is a like, if somebody said, you know, sometimes I will talk to people to say they're not technical, and I'm like, well, level of technology you need to master. You don't need to go back, necessarily, and get a PhD in computer science.

00;28;31;22 - 00;28;52;11
Jeff Nickerson
What you might need to do is you might need to talk to an agent say, hey, I need to learn how to run the API calls. Teach me enough Python to make API calls and make me and test me to see if I'm actually like understanding these concepts. So, you can do that. So, you can master, instead of trying to master everything,

00;28;52;14 - 00;28;59;16
Jeff Nickerson
you can master a particular aspect of this and then proceed.

00;28;59;19 - 00;29;23;17
Jie Ren
Okay, since we are talking about, like in the learning, for example, learning Python or learning technical stuff. I want to shift to our final question, which is, like both of us are educators, right? So, from an educator’s point of view, I know maybe not every single student wants to become an entrepreneur, or at that moment, at this very in this very life stage, maybe a few years later,

00;29;23;17 - 00;29;42;28
Jie Ren
and then that person is that students thinking about becoming an entrepreneur. So, for the students who are just graduating from these programs, right? How? What is your best suggestion for them to adapt to this workplace where, like AI, agents or personas could be used?

00;29;43;01 - 00;30;17;22
Jeff Nickerson
Right So, um, so I think there's, I think, like in everything else, experience is important, right? And in this case, the experience that’s going to be super important is experience working with AI Agents to build things. And so, my kind of general recommendation to students and people who have recently graduated is to you know, is to have you know, is to have subscriptions to you know, to the major models,

00;30;17;25 - 00;30;33;23
Jeff Nickerson
right, is to do that investment. So, you're using several of the major models. The reason to do that is they, they all have different qualities that are changing at different times. (Jie Ren: That’s true). And you can start to generalize. If you're working with more than one, you start to generalize, and you'll learn things. They’ll react in different ways,

00;30;33;23 - 00;30;57;01
Jeff Nickerson
and you'll just form a better image of what's going on. So, so for example, we have, we have open AI, we have anthropic with Claude, and we have Gemini. Gemini 3 just came out, which is, which is getting rave reviews at the same time (Jie Ren: Yes), so at least those three. And then if you're more technical, you might want to work with some of the Open Source Models and run something locally. 

00;30;57;01 - 00;31;17;07
Jeff Nickerson
(Jie Ren: Yes). Right? Because that way, when you run it locally, you can have it do things essentially for free, and you can let it do things over, like, much longer periods of time. So that'd be my suggestion, is to, is to have these things and keep on working with them, having conversations. And I don't even think it matters exactly the conversations that you're going to have.

00;31;17;09 - 00;31;39;21
Jeff Nickerson
You can have conversations that are like, you know, I plan to travel to a foreign city, and I want to photograph some architecture and where should I go, and then, and then do planning with it, and start to and even from that, that's like a, you know, that seems like it's, seems like it might be a frivolous use of the technology,

00;31;39;24 - 00;31;58;02
Jeff Nickerson
when you think about It, it's actually planning, right? And if you can start to get it to form a decent plan, it's not that far on to think about, you know, or think about a product, and the way you need to distribute a product, or you're a salesperson, and who should you be visiting when you go to a particular city?

00;31;58;02 - 00;32;13;17
Jeff Nickerson
These things are very intertwined. And so, I think, like in our everyday lives, using our everyday lives is actually not a bad way of getting experience with what they're good at and not good at. As you can tell from his recommendations. If you go to the city, you try to follow us recommendations. You can see what worked and what didn't work right,

00;32;13;17 - 00;32;17;24
Jeff Nickerson
what is it very sensitive to, and what is it less sensitive to?

00;32;17;27 - 00;32;31;04
Jie Ren
Thank you so much Jeff for this wonderful insight, I really enjoyed our conversation and I’m also looking forward to a new Artist in the future and hopefully you know you could be you know on the podcast again.

00;32;31;06 - 00;32;33;06
Jeff Nickerson
Thank you very much.

00;32;33;06 - 00;32;33;18
Jie Ren
Thank you.