The AI Training crisis: Why companies are spending money on AI but not educating

This is the Everyday AI Show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. This podcast is sponsored by Google. Hey folks, I'm Amar, product and design lead at Google DeepMind. Have you ever wanted to build an app for yourself, your friends, or finally launch that side project you've been dreaming about? Now you can bring any idea of life, no coding background required, with Gemini 3 in Google AI Studio. It's called vibe coding and we're making it dead simple. Just describe your app and Gemini will wire up the right models for you so you can focus on your creative vision. Head to ai.studio/build to create your first app. I literally can't tell you the number of times that I've talked to business leaders who have spent their companies anyways are spending usually millions of dollars on AI. Yet they haven't formally trained their people. And it's almost baffling to me, right? Because here we are with this generative AI technology powered by large language models, arguably some of the biggest technological shifts ever. And it changes almost daily. Yeah, why aren't companies investing in their people to make sure that they understand what the technology does, understand what it can do. And the cultural process changes needed to actually get a return on AI. So that's what we're going to be talking about today. Going over the AI training crisis and my companies are spending so much money on AI but not spending the time and the resources to educate their people. Alright, I'm excited for today's show. I hope you are too. What's going on? My name is Jordan Wilson. Welcome to everyday AI. This is your daily live stream podcast and free daily newsletter helping everyday business leaders like you and me not just keep up with the AI changes because they're happening literally every single day. But how we can make sense of them and grab the important insights to grow our companies and our career. So it starts here with the unedited unscripted live stream podcast but to take it on the next level. Make sure you go to our website your everyday AI.com there. Make sure you go sign up for the free daily newsletter. We're going to be recapping the highlights from today's podcast as well as all of the other daily AI news you need to get ahead. Alright, you don't got to listen to me rant about this one. I've done that enough. I'm excited for our guest for today. So live stream audience, please help me welcome to the show. Dan lawyer, the chief product officer at Lucid software. Dan, thank you so much for joining me every day. I show Jordan. Thank you. I'm thrilled to be here. I'm pretty excited for a chance to talk with you and to share some thoughts with you and your audience. Alright, so before we get into the topic and yeah, this is going to be a fun conversation. I think tell everyone a little bit if they're not aware. What does Lucid software do? I'd love to. Lucid software. We are a work acceleration and visual collaboration platform. It's used by more than 100 million people around the world. So a lot of people probably know about Lucid. You know, we're on this mission to help teams seem to build the future. And we do that to portfolio products. Things like Lucid char, which is in thousand diagraming. Lucid's far virtual whiteboarding, air focus and AI powered product management and real macro platform. So a suite of products that work together to just really help people solve some hard clasers problems. And and I'll ask you this and I'm sure we're going to get into a little a little bit. But even for you all personally, right? You know, why you said one of the larger companies in the world when it comes to putting AI products out there for people to use. What have you all even learned internally right when it comes to investing in AI products in AI offerings for yourself and for your customers yet training what's been some of your biggest takeaways internally. Yeah, there's there's a couple of things internally that we see one is it's it's actually much more of a cultural shift that it is just a retool of the team. And of course it's important to provide tools and provides space and time. But but it actually has to be treated like a cultural shift and an evolution of the culture of the company to be a company that embraces AI. Most how to use it has expectations and normal season even has like I think of them as cultural moments where AI comes to the forefront that highlights it for people and gives them permission and expectation and things like that. So the cultural experience has to be very well managed and distance like the security of the cases, the data, the tooling, the training that people talk about. I think that's the biggest surprise is how much of the culture impacted actually. So you have a deep background working in product at some large companies. So you know, I like asking people this right because I think sometimes you can learn through personal stories. I've shared mine plenty, but can you talk a little bit maybe about when was the first time or if you remember, you know, when was the first time that you looked at an AI system and you were like, wow, kind of taken it back. But almost like not taking it personally, but when was it at the point where you were like, okay, this piece of software or LLM just produce something that I didn't think it could and this is something maybe some knowledge that I thought I was kind of special at knowing something at this level. Do you remember that or, you know, do you have any anecdotes kind of like on that kind of point of realization? Yeah, well, like maybe I'll do that a two part and so the first time was actually the one coming to when I worked at ancestry.com and back then, like we would have probably just talked about machine learning and stuff like that. But it's actually what we're doing is closer to what we think about today is AI than that. And so like we like what we're able to do to like automatically generate stories and information about people's families and how the fun people was amazing, but you fast forward to like the more recent generative AI world things like that. I think the first time that I was able to go to an AI and something that lucid that built and and give it a prompt basically asking it to, you know, diagram out for a very complex system. And it did it and got it, you know, 95% right. I was like, okay, like, like that's pretty cool that like I can see how that could change and speed up the way that I work. Being able to just like the time to understanding side was someone's faster when I could do that. So I that was that was like a first lock me there's been many unlocks and how I think. So I kind of want to jump to the end here. So this big AI training crisis, because depending on what the, you know, the stat, the study that you look at, there's so many. But I say for across the board, most stats say that, you know, 90 plus percent, you know, of executives say that AI is a top priority. Right. And, you know, obviously the amount of money that companies are investing into AI, you know, it's, it's in the billions, right. Yet most studies show that only a third or less are properly training their employees on how to use it. Why? Why this big gap? Why is everyone saying this is the most important thing? And we'll gladly spend millions of dollars. Yet why are employees not getting trained? Yeah, I think there's there's several gaps in there. One of the gaps for employees not getting trained is it actually like takes a little bit time for companies to how do I safely provide access to the tools in a way that doesn't, you know, compromise or data doesn't compromise. There's been there was initially a lot of fear and concern about that that the concerns is still there, but there are a lot of playbooks now for how to like do that. So that part is accelerating. Then second part is like like what should I train them on? Like like there's a broad general training up just like, well, you give it prompt and you should answer and things like that. That actually doesn't take you very far and being able to get your work done. You have to go deeper and think about how to train in domain specific areas. You have to actually have a fair my understanding of the domain is so matter expertise to really extract the highest value and actually think one of the biggest gaps and how people think about, you know, getting value from AI. It's the combination of training but also the expectations are like in order to get a good out from from AI, you know, gender to AI is non deterministic businesses don't survive that very well. They need to, you know, predict outcomes. And so you have to teach people that to get good outcomes from AI to actually have to ground the AI and what a good job looks like. You have to ground the AI and, you know, to reality of like, this is how work gets done at our company if you want to automate that work. And so you needed like actually back people up and teach them, okay, you have to actually have a fair amount of documentation that you can provide to the AI about how your company works and about what a good job looks like before you can then get the highest value from AI. And so it takes some preparation and some fourth eye and something specific knowledge to be able to do well. Yeah. And analogy I love especially since, you know, I interviewed the guy who, you know, came up with the easy button right like way back at staples at HP now. You know, but it seems like that's the expectation that a lot of business leaders have. They're like, okay, well, you know, especially larger enterprises if they have tens of thousands of employees and they're like, all right, well, we'll pay the, you know, the 20 or $30 a month for tens of thousands of people, which adds up to, you know, usually seven plus figures annually. They're like, all right, well, there's the investment now. It's an easy button. That's wrong, right? Yeah, it takes more like like there is, there is a there, there is an outcome there. But it takes more for thought and preparation, maybe than people initially thought, and we think about this loses a lot. We talked about as the last mile problem, right, which is, there's, you know, if you think of logistics, right, you build a bunch of distribution centers. That doesn't matter actually unless you can get it from the distribution center of people's homes and similar to AI is it's like you can have the AI license the tool. But if you, if you can't actually, you know, pass to the AI information about how your company works, which requires you to actually go through and document your processes and document things and get the knowledge that's scattered across many people's heads and get it all together with people to see it. And then to make it worse, like if you pass the AI bad process, you'll still get a bad outcome, right? So you actually have to have to document how your company works and then you have to refine that. And that's part of the essential training is like you have to teach people that that part of what they need to do to get the most value from AI is to document how they work so that they can share that with AI to have good examples of good outcomes so they can share that with the AI. And so there's a fair amount of like teaching and expectations setting, I think that has to happen there and I'm glad you brought that up because I, I felt weird, you know, back in like early 2023 saying like, hey, you need to talk with an AI. You need to teach it. You need to train it just like you would with an employee because I think back then everyone was looking at large language models like chat GPT or or Gemini or Claude is co pilot input output. Right. And not necessarily a coworker yet. Here we are, you know, rolling into 2026. I think it's a little different now. I think people are looking at AI, especially agent to AI as true co workers, right. How can people get through that mindset shift of, hey, this is actually something I need to sit down with. I need to iterate like your example. I need to ground it not just in our company's data, but ground it in, you know, what good work looks like. How can people get to that shift because it is hard to treat a non human thing with a human ask characteristic of working with it being patient and sharing. Yeah, it even shows that like that you see these like, yeah, like we've done a bunch of surveys and you see gaps between like an executive or leaders mindset and how they interact with AI and how an individual contribute might interact with the AI and it really has to do with this, this comfort of delegating work. And this comfort of like having this award like, like, you know, I'm very used to asking other people to do things for me and expecting high outcomes from that. They're, they're, you know, I think if you take Dan 20 years ago, I didn't even think the same way. And so, so it takes some of that, but I think there's a, there's an evolved model that come to us actually how we tread like incorporate into our own products is, is like the idea of like AI as a co-collaborator, co-collaborator, not as a support. And I think many people probably feel more comfortable with that is like, he liked we've got, you know, a six man on the team now that we can turn to, we can trust who can get things done, but we still have to give them feedback like with any other team member. We have to do that. So I think, and I've even like, I like, I wouldn't go all the way there maybe, but mentally sometimes I like I tend to personify not my assistance quite a bit. I tend to talk to them as if they're real people, but then I have to back them off because then they tend to talk to me like, like, I worry that you know, like, like, AI assistants try to flatter me. And I have to tell them, I'm like, look, I don't want, I don't want you to flatter me. I want critical thinking like I don't, I don't need you to tell me this is good. But like I need honestly, so I actually have to have to say things to AI to get it to give me more critical of feedback. Otherwise is just telling me everything I do is great, which is not true. So there's like just like a whole work pattern that we have to things. Yeah, just just like humans, you know, the the AI is being a little too psychophantic to try to, you know, suck up to us. So you said something there that I want to dig a little bit deeper on just this concept of, you know, AI and maybe treating AI like a subordinate, but maybe that's not the best way in the future. But real quick, before we get into that, a quick word from our sponsors. Have you ever wanted to build an app for yourself? Your friends are finally launched that side project you've been dreaming about. Now you can bring any idea to life, no coding background required with Gemini 3 in Google AI Studio. I like what you were saying kind of about treating ai like a sixth man. For basketball fans or maybe for not a basketball fan. The sixth man is important. That's the person that comes in first in basketball and usually they can play a variety of positions. They're good enough to be a starter but for whatever reasons they're not a starter yet. What happens in 2026 Dan when the models themselves are all starter worthy right and maybe they're better than all of the starters. How do we number one get over that mindset shift right and I'm starting to see that a lot personally not just handing tasks off that I would like a subordinate but oh this is if I had someone running my own company right and I'm taking you know those big picture. You know answers or outputs from it and now I'm the subordinate are we getting to that point and if so how do we prepare for that and how do we trade for that. Yeah I think we are getting closer to that but it's actually really interesting because it's very similar to how you treat another person right you have to earn a certain amount of trust. And sometimes AI like and it's on a like a skill by skill or use case by use case basis you have to gain trust that it can do a consistently good job at something. And so part of how that trust can be built is you know to to inspect what it's doing is one of my pet peeves on my team with somebody hands me something that's a generated and I'm like did you even read this like like it's not quite right but so so you have to do that and and to increase the likelihood of growing that trust and get there. You know AI does so much better if it has like broad context and broad knowledge of the world but when you can grab a specific context about your business about your domain about what you do it'll do so much better and if you can keep feeding it it causes stream of content. Just like you would another person like team so so it stays very aware what's happening it's going to do better and better so I think you know you gain trust on a use case based basis that probably then starts to build awareness on the team like K would actually discovered that that if we you know ground AI with this knowledge of how our company works. And this particular use case it can give us a consistently good outcome and then that then inside you should share it across team and the team can start using that and get that in there but but I'm skeptical that that you'll ever get the strong outcomes you need without providing specific context to the AI. I want to go a little bit deeper and maybe this would be a little little technical and dorky but you know one thing we keep talking about here is is grounding and obviously that's extremely important right when working with non deterministic generative AI large language models right that are in theory just next token prediction right. But when you ground it right in your company's data and if your data is clean and if it's organized right i'll say in 2024 that's a big part of what humans did right they made sure to feed company information to a large language model right especially when we're talking about front end chatbots right so leaving the the API in the dev talk you know at the door here but on the front end now in 2025 all the major systems right they essentially have. You know two clicks and now these systems are grounded in your company's data whereas in 2024 I was a big part of what you know AI native organizations what the humans did there right they were just making sure so now that that's you know it's not solved sure right but grounding is relatively simple and straightforward now so moving forward into 2026 now that these large language models it is much easier to ground them in your data. How should we be thinking about working with large language models when they do have access to that and if your data is in a row right how does that change the role of your everyday business leader going forward yeah so there's still a missing piece right so you so you've got the data the data is not the workflow the data does not explain this is how you go from A to B to see the D to get the outcome that you want. And so you in addition to having the data you have to actually ground it in the process this is how work it's done this is how you know this is like I'll be really practical right so like how at your company do you reconcile a while transfer wire transfer. Right like data will not tell you that and but and in fact there's probably not a single person your company will tell you probably have to get 10 people together to answer that question but if you can document that and pass that to AI then the combination though this is this is procedurally how we work how you get it down this is the data set and this would a good job looks like then you can do it so that there's actually a missing component beyond just grounding the data is going to get in and the process carrying in procedural knowledge. Of of how to get things done and you can imagine world where you have like mtp servers and strong API between all types of systems you still you still have to have an orchestration that says this is how to progress the work this is the proper sequence now I think over time the I can be trained and it can learn that but there will still be like specific knowledge in a company that is that is their secret sauce that well we're better because we do this way and we don't want the world to know that. And so I think for a long times can be important for companies to to augment the data procedural knowledge that this is how work is done and then you'll get a good outcomes. You know Dan I think you've been spot on just the amount of times that you've mentioned documentation process documentation and having your data in order I think those are two keys that you know you can't overlook but one of the biggest issues I think you know when it comes to educating employees is the right of change right if we just look at from you know mid November until now every single week it started with open AI and then it went Google and then it went rock and then it went Claude and now we're back to open AI releasing a new best model week over week we've had it for five straight weeks now. So you know especially when companies are maybe using one or two systems and they're changing all the time and you know a lot of people don't they go and you go from a gpt five one to a gpt five to all I can just do things the same well you can't always how can companies possibly keep up when the technology they're using and maybe everything they've learned can change very quickly without very much notice. I think there's two critical components that one is you have to think through how do you make it easy for your employees to rapidly access the new technology in a safe way. And so like it is like you know what is your procurement process what is your policy around what I can install on my machine how can I do that safely and quickly like like you have to have like like you almost have to have like here's a fast path. And these are you know the guidelines of how you can do fast path experimentation you can't use customer data you can't use P.I. I can't you know like that kind of stuff but we do need you to rapidly experiment with new things and then you know so you need like those fast paths of how you can you get people exposure to the new things that are coming you need to provide them time and space and cultural moments to highlight it. So that's the one piece to have the other pieces you have to know what you're getting by from and so you have to figure out for like every kind of business function or man we're trying to get by for may I what like what is the central measure of speed. You know how to for example like and I you know I'm attentive like I tell you our measures but like like you like you think of like a software team. How is there a number that measures whether A.I. is speeding you up that is beyond just like you know the sentiment but like an actual you know quantitative view of is this new tool speeding up or on a product you like product you I see like that's my way where it's been most of my time. Like how do I measure if A.I. is actually making us faster at getting good outcomes and we we spent a lot of time figuring that out for our company like what what are those quantitative measures it's so so that gives you like a way to evaluate and say is it is it just new and shiny or is it actually creating value and speeding us up toward better faster outcomes for our customers. And so I think that's the combination right is like allow for rapid access and experimentation but have a quantifiable way of knowing if it's healthy. So rapid experimentation what one of my favorite things right yeah don't don't don't spend you know hours or days or weeks on something if you aren't ready for plan B plan C or experimenting with them at the same time. Right and Dan I think what you said there just about having those kind of you know internal benchmarks and quantitative measurements extremely important especially when you're working on something a little bit more finite right but what about for everyone else what if there is no one benchmark if there is no one measurements you know on one system to see if there's. You know a good return you know for maybe for those that are looking at training their entire company and maybe this is something that you've all learned internally maybe what's been some of the most successful ways that you've seen even internally on hey here's good ways that we can educate our people in space that is changing weekly. Yeah so I think you know we create cultural moments that are like and I think there's like you know so there's things like like all hands meetings or staff meetings as things like that and we create space and all those places to highlight what we're learning about AI right so like in my you know all hands meeting I have a you know an AI moment where we're we're having people highlight various ways that they're seeing new value or new experiments what's working what's not working around AI. To share knowledge probably because you have many people touching them and and and we need the knowledge we leverage and so so creating those cultural moments it doesn't include one that shares the knowledge to it gives people permission to play and and it it's highlighting hey this is a good job this is somebody who went and tried something new with AI and it did it did not work but we're we're like you know giving them air time and and highlighting that's what a good job is is to go do this type of experimentation at the same way. And so I think like like that is a critical thing and and and pretty much any part of any company can figure out like where are the cultural moments where we give the air time to AI to to like start working on the behavioral change and help people realize that it's safe to play now it also have to create the space right like expecting people to just like go home and you know spend their after hours doing all the learning like some people will do that. But I think you have to give them time space at work to do that and you know so like you can you take hackathon style approaches or can you say we're going to have have the Fridays where you're free to experiment or things like that so that you know it'll be different for every company house but it should be I think intentional how they do that. So we covered a lot in today's conversation you know everything from going over the you know cultural changes and talking about data and process documentation and having the right quantitative measurements internally so you can know even what to educate people on. But for those business leaders right now who are planning out there 2026 AI education how they're going to get it done what's your one most important piece of advice for them to get education right in 2026. So so one think of the prep work that has to be done to educate and like there's there's like layers education there's the broad general AI awareness and the real value will come when you get domain specific and talk about like in this domain in this part of my company for this business function. This is best with leverage AI and then a part of that training is in order to get the highest values you have to ground the AI in both the data and the procedural knowledge of how things work you'll get better outcomes. So it's like it's it's you know getting from general to domain specific to very pragmatic around how to get the highest returns from AI will help and then creating a culture around that that reinforces and supports. The idea of learning and training and rapid experimentation. All right some great pieces of advice as it's you know big topic we're all trying to tackle and Dan your time to get today I think helped us you know tackle this thing a little bit better so Dan thank you so much for taking time out of your day to join every day AI we really appreciate it. Of course thanks Jordan have a good one all right and if you missed anything y'all don't worry we're going to be recapping it all in today's newsletter so if you haven't already make sure to go to your everyday AI.com sign up for the free daily newsletter thanks for tuning in we'll see you back tomorrow and every day for more every day AI. Thanks y'all. And we'll help you get started head to AI dot studio slash build to create your first app. And that's a wrap for today's edition of every day AI thanks for joining us. If you enjoyed this episode please subscribe and leave us a rating it helps keep us going for a little more AI magic visit your everyday AI dot com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.