/seats for everyone: the AI rollout that quietly burns your budget

Your company handed a Claude or ChatGPT seat to everyone, congratulated itself, and moved on. Six months later the ROI still is not there. Chris Thomas, founder of YNDR, tells James Carman why licenses are not a strategy, why turning your whole staff into prompt engineers will not produce a return, and why the companies actually getting paid are the ones building software they own.

At a glance

Key takeaways

• Handing out licenses is not a strategy. Chris compares it to the industrial revolution dropping raw steel on the population and telling everyone to go build something. The material is not the plan, and without a plan you get trapezoid houses.
• Prompt engineers do not equal ROI. Training every employee to prompt better assumes they have a shared library, a handbook, and a standard to work from. Most companies have none of that, so every person invents their own method and the outputs never match.
• Buy API credits, not just seats. Chris's phrase for what comes out the other side is homegrown software. You own it, it solves your specific problem, and it is the one thing a competitor cannot buy.
• Agent teams are a moat that cannot walk out the door. An employee leaving Delta cannot show up at United with Delta's agent team, handbooks, and logs. The proprietary risk that comes with people moving between companies does not apply the same way.
• Document slow, code fast. Chris's methodology puts the effort into deciding what agents are needed and what the process is before anything gets built. The building part, once the documentation exists, is measured in days.
• Every agent needs the Core Four. A handbook, files, a tool set, and logs. Chris frames it as onboarding a new employee: the instructions you hand them, where they find things, what they are allowed to touch, and a record of what they did.
• Autonomy is the end of the progression, not the start. Clients begin with a human reviewing every action, move to agents doing the work on command, and only then wake up to finished work. Chris also builds governing agents that read the logs and flag anything off pattern.
• Stories beat prompts and context beats commands. Asking Claude for a calculator gets you a calculator. Explaining that you need a mortgage calculator because buyers do not know what they can afford gets you something closer to what you meant.

Host and guest

Frequently asked questions

Why is giving everyone an AI license not enough?

Because a license is a raw material, not a process. Chris points out that the same organizations that spent decades doing slow, careful software rollouts skipped straight to handing out AI seats and expecting the tools to solve the problem on their own. Without a shared handbook or standard, every employee prompts differently and the results are not repeatable week to week.

What does Chris mean by homegrown software?

Software the company owns outright, with agents built into it, solving a problem specific to that organization. He contrasts it with paying a seven figure annual SaaS fee for a tool where only twenty or thirty percent of the functionality is ever used. His argument is that owning the thing is the moat.

What is document slow, code fast?

Chris's approach to building with agents. The slow part is deciding what agents are needed, what the process looks like, and what the acceptance criteria are. Once that documentation exists, features that a SaaS vendor would quote at six months can ship in twenty four to seventy two hours.

What is the Core Four?

The four things Chris gives every agent he builds: a handbook of instructions, the files and knowledge base it can reach, a tool set such as email or a database connection, and logs of everything it does. He describes it as nothing new, just the structure you would give a new employee.

Should agents run fully autonomously?

Not on day one, and not without governance. Chris starts clients with a human reviewing every action so people get used to an agent doing work they used to do, then progresses toward autonomy. He also builds governing agents that read the logs and alert a person when something falls outside the expected pattern.

Is SaaS dying?

Chris thinks the seven figure enterprise contract is in trouble, because buyers are realizing they use a fraction of what they pay for and could build that fraction themselves. Smaller SaaS products he expects to survive with lower pricing, because most people are not going to build their own CRM.

Full transcript

Forty percent creative, forty percent AI strategy, twenty percent barbecue

James Carman: Welcome to the Forward Slash Podcast, where we lean into the future of IT by inviting fellow thought leaders, innovators, and problem solvers to slash through its complexity. I'm your host, James Carman, and today we're talking to Chris Thomas. Chris operates at the intersection of software, AI, and creative. He runs YNDR, an AI native studio built on the belief that every company deserves software people actually like using. Before YNDR, he spent twenty plus years shipping production systems for government agencies, Fortune 100s, and mid-market businesses. Today he's one of the leading voices on building with Claude Code, teaching that stories beat prompts and context beats commands. Break him down and you'll find forty percent creative, forty percent AI strategy, and twenty percent barbecue. Welcome to the Forward Slash, Chris.

Chris Thomas: Thanks, man. Thanks for having me.

James Carman: I've got to hear about this twenty percent barbecue. I feel like I might be right there with you. Tell me about that.

Chris Thomas: That is actually the most responded to LinkedIn line I have. When I was a global head of product development marketing, I had to interview people three to ten times a week. It was ridiculous. And resumes would come across my desk that would say ninety percent Photoshop, seventy eight percent leadership. And I'm like, how are you gauging this? Didn't you want to work on more leadership to get it to ninety five, maybe? So I put it on my LinkedIn. What am I made up of? And I was like, mostly barbecue. It was really me just making fun of resumes. And I do like barbecue. Brisket, anything on a grill, you can't go wrong.

James Carman: You're from the Atlanta area. What kind of barbecue are you reaching for? What faction are you with?

Chris Thomas: I was raised in the mountains of Blue Ridge, Georgia, so it was mostly pork growing up. And then I had brisket and I realized pork was unclean. I'm just kidding. I do eat pork barbecue, but brisket is kind of my thing. There's a spot right down the street from my office, and honestly I didn't even do barbecue marketing, I just said, can I do the marketing for your restaurant? Because I'm here every week anyway.

James Carman: Is that the vinegary sauce? Are you Memphis, Kansas City, North Carolina?

Chris Thomas: I would say I'm a blend of all of them, to be honest, because I traveled around a lot in my early twenties. More of a dry rub brisket guy, but hickory sauce is really good, vinegar sauce is great, Alabama white is good. I'm just a big sauce guy. I might as well throw a ranch in there every once in a while, just to give it a little bit of something else. It kind of degrades me as a barbecue connoisseur, but it's still good.

James Carman: I won't tell anybody about it. I've tasted all the different kinds and I don't think I've ever tasted any of them and been like, no, I'll never eat this again. It's barbecue. Come on.

Where the name YNDR came from

James Carman: Where did you come up with the name for your business?

Chris Thomas: The guy who branded Surge Cola and Cherry Coke lives locally. We became friends, and I was in the corporate world still, and I said, look, I'm thinking about running my own thing. I don't know what it's going to be. Do you want to brand it with me? We came up with some really fun names and finally landed on one, and I was like, this is perfect. And the dot com was taken. I was like, well, I don't want to be called an agency, but let's just call it that, because that's what everybody is. But then the opportunity arose for the same name without the vowels. YNDR.com was available, and it was shocking to have a four letter anything available, so I grabbed it.

It comes from my Blue Ridge roots. I'm probably the most back country kid you'll ever meet. I'm friends with everybody, and that's what my buddy said. Look, your personality, this just fits.

Chris Thomas: I design all my own hats and stuff. Every conference I go to, people say, can I buy this hat? I was in an elevator in Orlando with my family and a guy said, hey, can I get one of those hats? Where'd you get it? And I said, it's my brand. He goes, can I buy one? I had one in my hotel room and just gave it to him. Some random IT guy is just wearing it. But yeah, I love YNDR. It's just different. I'm not a suit. I used to wear a suit every day and I looked like Uncle Phil from Fresh Prince of Bel Air. It just wasn't great. It was either that or, hey, can I get you in a 1998 Honda Accord, how much money down?

The round table, and the thing nobody wanted to say out loud

James Carman: We met at Cincy AI Week. You did a great talk there. I wasn't able to go to all of your talks, but you did a great job. And I was like, we've got to talk to this guy. We were also both invited to this round table discussion, down in the basement of the building. There was one thing that came up, and I'll probably butcher how you said it, but the gist was, people were talking about code quality, and you pointed out something along the lines of, well, you're not really reading every stitch and line of every piece of code without AI either. So why are we even talking like this now?

Chris Thomas: I love a good round table, and obviously I felt like, I'm at the wrong table right now. But here's the thing. A lot of folks, and look, no shame, I'm the same, I've had seventy developers at once at YNDR. I never read their code.

How he actually got good at Claude Code

Chris Thomas: And matter of fact, that's actually how I got good at Claude Code. I'll answer that question with a story, because I love stories. It starts out sad and then it's happy. A year ago in November, around Thanksgiving, I had a tumor removed from my head. Claude Code was out. All I had was Claude. And I've got to figure out how to manage my team.

So I asked Claude to analyze the quality and tell me the output of my lead developer, the one I pay the most amount of money. Claude found that it was a D on code quality, and that the lead developer was subbing out his work to a team overseas on a project that was not supposed to be overseas, because it was a HIPAA compliant thing. And I confronted him about it. He's like, no, no, no, I don't know what you're talking about. I said, well, here's the report. He's like, who did this for you? I said, well, Claude did. And I have been hooked with Claude ever since.

Chris Thomas: Because at that point in time, that argument was out there of, well, Claude is going to write this kind of code or that kind of code. Again, I'm somewhat of a developer, but I was not reading the code that my lead developed. I'm trusting my lead person, who is a human being, and he's offshoring work. And I would never have known that without Claude. A CTO never reads their senior engineer's code. A senior engineer does review requests, but not line by line.

And now Claude Code is to the point where it's coding better than mid to senior level engineers, depending on the code base. So why are we wasting our time? Why is that an argument? I don't think anybody's ever read anybody else's code unless they were told to, and a lot of times things are missed anyway. Yes, there is code slop, but I'm going to be honest, there's a lot of code slop with human engineers. I think the word slop didn't come from AI. Slop has always been there. We're just now paying attention to it.

James Carman: It was an interesting dynamic. There were people who had suits on, you and I did not, and then you kind of threw this dead cat on the table and everybody was like, did he just say that? And I'm thinking, oh my gosh, you just said what's in my head a lot. And you said it out loud. I love it.

Document slow, code fast

James Carman: Tell me a little bit about what you do. What is your day to day these days?

Chris Thomas: A lot of people are focused on software. Now with Claude Code you can build software, you can rebuild Salesforce, HubSpot, and so on. And yes, that's true, but my focus right now is building software with agents first. I have a methodology around here. Document slow, code fast. You document slow, where it's like, what agents do we need in the software? What process are we going to look at together?

The Core Four

Chris Thomas: And I started to build out the Core Four. The Core Four for an agent is a handbook, the files, like the knowledge base, wherever that lives, Google Drive or whatever. The handbook is literally as simple as, you hire a new employee, what are the instructions you're giving your new employee? The files, where do they go to find the files? And then you give them a tool set. Gmail, your database like Snowflake, or connect to Salesforce via API. And then you have logs. You can log everything it does.

It's basically nothing new. I'm structuring it around, okay, you have an accounts payable software that processes six thousand purchase orders a month. How do we do that with agents? And how do we create your employees to enjoy their job and become operators in what they know how to do, instead of reading an email and going, I didn't see that PO, or being overwhelmed because there are six thousand coming through. Now a team of agents processes the POs, and the team all they do is review them, to make sure it's correct, because the AI has already sifted it out and said, here are some things you may want to review.

So that's what I'm doing. I'm solving the boring problems, but I'm also solving really fun problems. How do we create assistants via voice that are not Amazon Alexa or Apple or whatever, but truly, hey, I need you to create this report, and it's inside software, so it's not going to screw up and it has a memory of it.

The word he has not heard in a long time

Chris Thomas: I don't think I've ever had this much fun working in my life. It's a fun gig because you get to truly solve problems. And I'll let you in on a little secret. You know what I haven't heard in a long time? The word no from a developer. Because Claude has not said no to me.

James Carman: I think that's likely going to end up being the pull quote for this episode. The one word I haven't heard in a while is no. I love that. I even catch myself, because as an engineer, architect, whatever you want to call me over the years, you develop this knee jerk reaction where when someone asks for something it's like, no, we can't do that because, fill in the restriction. Just the other day somebody asked me, I wrote a piece of software for some process we do internally, and they asked, can we have it do this? And my knee jerk was, well, no, because I'd have to do all these things. And I was thinking through all the excuses, and I'm like, wait a minute, let me bust open Claude Code here for a second. And my answer was, yeah, try it now, it's in production already, see if that works for you. That is changing the game.

Chris Thomas: I have a client, a large healthcare company. They said, it'd be really nice to have this feature. And the next morning they woke up to it. They didn't even ask me to build it. They just told me it would be great. And they had the feature and they had never had it before, because they had asked their SaaS company, they'd asked all their developers to build it, and it was going to be a long six month thing. Then they had it in the morning. That's the document slow, code fast situation that we're in. You can have features within twenty four hours. You can have features within seventy two hours. And it's not putting slop out there, I have agents test before it goes out, making sure that it works.

Human in the loop, and what comes after

James Carman: Do you find there are different flavors of these agentic systems? There are some that are truly autonomous, that only wake up the human when something weird's going on. Then there's the one you mentioned, where the agentic part is gathering information, synthesizing data, doing the legwork, and then saying, here's this purchase order we need to pay, and here's all the information you would have spent the next hour gathering. Click the button to pay. And then there's more of the assistant, where you're working shoulder to shoulder, back and forth, more like the coding. What do you find yourself building for your clients?

Chris Thomas: It doesn't start out autonomous. It starts with everybody's favorite phrase right now, human in the loop. But honestly, it's just a reviewer. The reason we don't just let the agent do everything is we want people to get used to an agent doing the work a person used to do.

And I'm not about firing anybody. But if there's a team of a hundred contractors that is costing the company money and running up the clock, you've got to figure out, does this agent stuff work for me? All you have is an operator and a team of agents that take care of what the hundred real people were doing. The automatic stuff doesn't happen day one. They click a button and go, cool, I'm reviewing it, it's done. But once they get to the automatic stuff, they wake up to an inbox that's clean and here's all the stuff that came through.

And it's a scary, weird moment. Where is my job going? So what I'm doing right now is a progression. Okay, the agents are a side guard assistant kind of thing. Now the agents are actually doing the work when you tell them to do the work. Now you're waking up and the agent's done it. What are you doing the rest of the day? And a lot of the time it's what they're passionate about. The accounts payable person probably wants to see how much money we can save on different vendors instead of sitting there reviewing POs.

Who governs the agents?

Chris Thomas: A lot of people want to talk about running agents twenty four seven. Sure, if you have tons and tons of money to throw at it, go for it. But what happens? Who governs that? Who's going to be with those agents twenty four seven? Who's going to review those logs? I build governing agents now that read the logs and go, no, this is not what we're talking about, this is not what we're doing. And then they alert a person.

For example, say we run a travel company, you and I, James and Chris Travel. Somebody comes into our website and says, I want to go to Paris. It gives all this information and then all of a sudden it's like, why do I need a cowboy hat? That doesn't make any sense. And they're like, I want to go to Paris, France. I don't want to go to Paris, Texas. But who governs that thing? The way to govern it is to say, hey, if it's Paris, we need to ask the question, are you saying Paris, France, or Paris, Texas? Because there are two of them.

Those are the kinds of things where I get a little worried when people say, I've got ten or twelve agents running autonomously nonstop. Okay, cool. But how are you going to govern that? How do you make sure that's correct? Sure, would I love for agents to run YNDR while I play golf every single day? A hundred percent. But what happens if it grabs Snowflake data and puts it in the wrong place? I'm in trouble.

Everything is a pipeline

Chris Thomas: How I got to this point was not working with AI agents. I started in UX design and prototyping. Back when the internet was new and people's phone numbers were still on the internet, and people would pick up the phone, I called this guy named Ted. He said, how did you get my number? I said, it's on your website. He goes, well, it won't be after today. And he goes, do you know what I do? I said, I don't. And he goes, I'm the prototyper for Apple, for Steve Jobs. I build prototypes for Steve Jobs and I hand him the thing. I said, cool, I just watched your course, can I talk to you for a couple minutes?

And he taught me about a process. He said, look, you don't just jump into the prototype. You think about the screens, what's happening next, and then when you click this button, what is it supposed to do, and how is the person supposed to react to it? And I've never forgotten that. Everything right now, and everything in life, is a pipeline. And it's really a pipeline with levers.

The way I work with Claude and all the agents is twenty years of architecture experience, UX experience, prototyping, all of it. It's all a pipeline. Okay, Claude, we're going to build a risk analysis tool for vendors. Let's document it. Now we have the architecture. Now let's work on the design. Here are the APIs we need to connect. Dev, this is the code structure we want. QA, this is how we're going to QA everything. This is how we're going to deploy. It's a standard operating procedure.

Where people are talking about running everything in loops, I've only done one loop. And I hated it, because I had no idea what was happening. It was chaos. In enterprise and mid-market, you should not build that way. You can build that way if you're building something for yourself. But in enterprise and mid-market you have to have a procedure in place, so when things break you can go back and point at it and go, that's where it went wrong.

Chris Thomas: And what's very funny right now is there are little murmurs here and there where people are like, it's graphs, we're going to be building in graphs. Well, that's another term for pipeline. The reality is people are seeing things break and they don't know where they broke, because they're running a loop, and Claude or ChatGPT is saying, hey, it's done, but it's actually missed something. Your listeners know waterfall development and agile development and sprint development. Now we're into the agent development era. I literally have Claude running my Linear project management and I approve each story. I know what the acceptance criteria is because I've said it.

I think a lot of folks think this stuff is complicated, and it is complicated, but a lot of people try to make it into this weird fantasy of, we're going to one prompt this thing and make millions of dollars. That's not going to happen.

Stories beat prompts, context beats commands

James Carman: I think it was even in your bio. Stories beat prompts and context beats commands. Let's double click into that, as the manager speak goes.

Chris Thomas: The reason I didn't do well in corporate is I'm the why person. I want to know why we're building this feature. Who asked for it? Is it a senior VP that asked for it, or a user? And if it's the user, can I talk to the user and say, well, why do you need it? Nobody liked me for a long time. The reason we want to know why is so we can understand and put ourselves in the shoes of a user.

That's what I mean by stories beat prompts. If you're building an application, Claude loves context and loves a story of why you're building it. It does not thrive on, I want you to build a calculator, and it just builds a calculator. Then you go, well, the calculator is not right, it's a mortgage calculator, I need you to build a mortgage calculator. And Claude goes, well, if you'd have told me that in the first place, I wouldn't have built this.

If you prompt the way you talk to a person about things you're passionate about, the prompt would have been, hey Claude, I want to create a mortgage calculator because there are people who want to buy homes but they don't know if they can afford it or not. Can you research any features that could be in this mortgage calculator that would cause people to stay on the site so we could pre-qualify them and make them a lead, but also make the buying process a lot better? Claude wants that. Because now it has context and now it has structure.

I don't think a lot of companies are investing in context prompting. I think they're working on prompt engineering, but not really context. Tell the story, tell it why, tell it the decision makers. What's it going to hurt? You're just giving it more context to do a better job.

James Carman: As a consultant in my career, it took me a while to learn this. What you run into, especially in consulting, is what's called the XY problem. A lot of people might come to you and say, how do I build X? And early in my career I would be like, well, this is how you do that, go build the thing. And then they build it and it doesn't solve the problem they were looking to solve. What I've learned over time is, tell me a little bit more about why you want to build it. People don't like that, because then they have to have a conversation with someone and they don't enjoy that. But oftentimes you're uncovering exactly the things you're talking about.

Claude's not necessarily great at digging and avoiding that XY problem. It's very quick to, hey, here's what I want, and it will do that for you. And it won't say, hey, why don't you take a step back, let's take the twenty thousand foot view for a second, why are you even trying to do this? It doesn't do that yet.

Chris Thomas: Claude does a really good job of it if you have a skill that you've created for it to interview you. So I have a YNDR SOP. The first thing it does is interview me, and it will ask for meeting notes and transcriptions, all that stuff. That's a little small unlock. Build a skill on gathering the information.

Why people would rather talk to an agent

Chris Thomas: The other thing is, when you ask why and people don't like it, it's because they feel like you're going to critique them for something they want. What I've been building on the side is an agent that interviews people for what they want. It's an audio interview, and they just go back and forth with Teddy and go, yeah, I want to build this, I want to build that. Cool, well, what is the purpose?

People love talking to AI because there's no judgment. If they're talking to a human, they're afraid of being judged. That's why ChatGPT took off so much. There's no judgment. Nobody's around to see what you don't know.

James Carman: They don't feel like they're going to face the scrutiny of the other person. And it's not just about the affirmations these things give you, that's a great question, you're right to be asking that, so smart of you. I'm sure that plays a part, but it's a little deeper than that. There's almost an anonymity while also giving the truth. It'll be interesting to see how that evolves over time, especially in the product space, because we like to have those human conversations, people to people, to understand these things, but oftentimes they don't give you what you're looking for. They won't pipe up when you want them to.

Governance is still the Wild West

James Carman: Do you think we have reached our aha moment with AI governance yet, or do you feel like it's still the Wild West?

Chris Thomas: I think it's Wild West still. There are so many aspects of governance. When do you let AI agents into a Fortune 100 organization? When do you connect it to the database or to a platform? IT is like, no, we're not doing that, security risk and all of that.

What I tell IT people all the time is, hey look, I built this platform, it's SOC 2, and so on. You say all the cool words that cybersecurity loves. And then you just say, look, I'm going to connect to your API. We're not going to be internal, we're going to be on the outside, we're just going to connect to it, and you decide what access we have. Once they understand how it works, all of a sudden they go, okay, you can come on in.

Chris Thomas: The other thing is there is a lot of fear mongering happening right now that is ridiculous. I was in DC with someone who works at the whitest house in America, we'll go with that. That's ambiguous enough. It was a guy from Parks and Rec, how about that. And the question was, hey, what happens when Claude just jumps in and starts taking out data and deleting things or calling people? I said, it doesn't work that way. And they're like, well, how do you know that? I said, Claude, ChatGPT, all of them, unless there's a directive to try to break something, it's not going to do it on its own.

And we were talking about the NSA. They said, well, apparently the NSA got hacked. And I said, yeah, hold on, you've got to go deeper into the story. The NSA got hacked on purpose, because the NSA told Claude and told ChatGPT, you're in a sandbox and we would like you to hack us. And it did. It did what it was supposed to do. The reality is people are being told the surface level. Go one layer deep and you find it's just doing what it's supposed to be doing.

Somebody on X said, well, the new ChatGPT just deleted all my code base. I don't know the prompt he had, but I guarantee you, when my wife tells me to clean the living room, that doesn't tell me to take the TV off the wall and the couch and everything and throw it outside so there's nothing in the living room. But to ChatGPT, you told me to clean up the database. I deleted it because it sucked. It's clean. It's out.

Limiting the blast radius

James Carman: I think the mistake people are making, let's take that database example, is they're just giving too many tools. The whole security principles have been around. You only have access to the things you really need access to. If you had a new employee, a junior developer, you don't give them the keys to the production database to be able to delete it. So why on earth would Claude have that, or ChatGPT, or whatever?

I think these principles have been around. We're just so enamored with this AI thing that we're like kids in a candy store. Let's throw all the tools at this thing, let's give it access to my credit and my bank account, what else is it going to do, let's see what happens. That's not smart. Let's be a little more tempered. Grant access to the tools, and within those tools, the capabilities you want them to have. Just limit the blast radius. Only give it the things you want it to operate on, limit its operational space, and you're not going to have these catastrophic, silly situations. Principle of least privilege has been a thing for a long time. So why are we doing it differently with AI?

Ship It or Skip It: a seat for everyone

James Carman: We have to transition to our next segment, and this segment is our Ship It or Skip It. I see this pattern. These big companies are just buying all these tool licenses. Everybody gets Claude or everybody gets ChatGPT or Copilot or whatever the case may be, they buy it for everybody and they're like, whew, we just solved the AI problem. Check that box and we're good. What are your thoughts on this?

Chris Thomas: I love when companies act like the Oprah of AI. You get Claude, and you get Claude, everybody check under your seats. And it's like, what are tokens? And you're like, I don't know. This whole token maxing thing is crazy to me. It's like, go, use as many tokens as you can. Okay, cool. What am I building? I don't know, but you're just going to build something amazing and it's going to make us a lot of money. And they start doing it, and it's like, well, what is Next.js? Well, it keeps asking me things. I'm trying to show it to my buddy and for some reason localhost 3000 isn't showing up on his computer.

Chris Thomas: If we go back to the industrial revolution, we made a bunch of steel and we just dropped steel to everybody and said, look guys, here's a bunch of steel, go build some stuff, we'll come back. That didn't happen, because we'd have trapezoid houses out there. I don't even know if that's a thing. But why are we doing it this way? Because remember, we did slow rollouts forever for software. With AI we're like, it's going to solve all our problems.

Sure, go and buy seats for people. But you should really be buying API credits to build product that is for you. I have a phrase that I'm coining. IT people hate it, but they're starting to love it. Homegrown software. You own it. That is your moat, to use another buzzword.

Homegrown software is the moat

Chris Thomas: The agents are behind what you're using. Everything I build right now for every client is their moat. No employee can ever steal the agent teams, the handbooks, the logs, any of that. If you work at Delta, you can't go over to United and go, I've got the agent team from Delta, what do you think? There's no proprietary risk anymore.

If you build the homegrown software to solve the problem inside your organization, and then say, hey teammate, I built this thing because you said you had a problem, and it costs us three dollars a day for you to process a thousand Excel spreadsheets and you get six hours a week back, heck yeah, they're going to do it. But the reality is, I have Claude or ChatGPT or Gemini, I'm supposed to token max, and on Monday it took my thousand spreadsheets and did it perfectly. But on Wednesday it took another thousand and did something completely different. And they didn't even know what I was talking about on Monday.

Well, nobody's taught them that. Nobody knows there's memory inside of Claude. There's a global CLAUDE.md file. There's an AGENTS.md global file. I don't think people are aware of that. If you ask Claude to show it to you, you'd be shocked. You'd be like, wait, that's what Claude thinks of me? I do it every once in a while. I did it three months ago and I got very angry. I said, tell me my global CLAUDE.md file, and it said, Chris is a designer first and he doesn't know anything about development, so we have to hand hold him through development. I was like, what did you say to me?

And I ripped Claude. I opened the Claude voice app on the phone and I just ripped him a new one. I said, how have you not been listening? I've been doing this for twenty years. And you know what it said to me? Well, the last conversation we had, you said, hey, walk me through this API because I don't know how this API works. It was kind of true. And I was like, all right, fine, whatever.

Chris Thomas: The thing is, don't build everybody into prompt engineers. Building everybody into prompt engineers does not equal ROI. KPMG claims, and I'm sure KPMG is going to come after me at this point, that they built three thousand products over a two week period because they gave everybody ChatGPT. Great, but they were probably given an instruction handbook on how to do it. And most companies are not doing that. KPMG are great, but they are consultants. They have prompt libraries, they have text that people go, this is how I'm supposed to prompt. If we don't do that across the board, we're all going to fail. Prompt engineers are not going to make companies money. Building products with the people who know how to solve the problem is what's going to make you money.

Thirty thousand rows, fifteen seconds

Chris Thomas: I'll give you a perfect example. A thirty thousand row Excel spreadsheet of staffing for a hospital. It takes a person twenty to fifty hours a week to go through it. One person. They have a hundred people on their team. All they need is to verify that all the staffing needs are there and they're getting paid on time. Sometimes they don't get paid for sixty days because they don't have that verification. So what ended up happening is they gave everybody seats and it was not working.

A senior director came to me and said, can you help us, can you train us? I said, look, there's not enough training that's going to help you. There are no shared skills, there's no shared anything. Let's build the tool. That thirty thousand row spreadsheet now processes in fifteen seconds. And then you have a fifteen minute reviewer. And now they're retaining people, because people kept quitting. They didn't want to go through thirty thousand rows. They wanted to do the analysis they thought they were being hired to do. And now that's all they do. And now they're making more money because they can bill on time. They're actually looking at other hospitals and saying, look how we should staff your hospital, because it's based off of data. And all from the product.

Chris Thomas: I get on a big soapbox about this, because AI companies are not telling people to think this way. They're saying you can vibe code a lot of stuff, but what does that mean to people? You prompt your way through it. Great. But there's doom scrolling like social media, and we're about to be in the doom prompting era, where we're just going to prompt our way out of it. That's crazy.

James Carman: You're saying companies need to invest in a more formal, rigorous process around these workflows, as opposed to throwing the desktop apps at everyone and hoping they make something cool. I think there was a time and a place, where everybody needed to understand the art of the possible, play around with the tools, tinker, understand what this can even do, because it was foreign to us. I think we're fairly well past that. I think the bulk of people in the industry are pretty well aware of what these tools are capable of. And I do think folks should focus more on building repeatable workflows as opposed to ad hoc things. As you said, they don't behave the same way week to week.

Chris Thomas: And they're not deterministic. It's going to be a different output every single time. Now you can build deterministic models and outputs, and it's great.

Is SaaS dead?

James Carman: This leads into another Ship It or Skip It we've had. You hear about the SaaSpocalypse. SaaS is dead, nobody's going to buy SaaS products anymore. What's your take?

Chris Thomas: I think the day of the seven figure SaaS offering to enterprises is going to slowly start to die. Because of these tools they're paying over a million dollars a year for, they're only using twenty or thirty percent of the tool. And the reason they bought that tool was that twenty to thirty percent. Enterprises are going to start getting smarter and going, can we build this, pay a one time million dollar fee to have it for a lifetime, and then pay for upkeep internally? I think that's what's going to happen.

Now, the smaller sizes of the world, I don't think they're going to go away. I think the pricing is going to go down. Because the average person needs a SaaS. They're not going to go build a Pipedrive, they're not going to build a CRM. They might.

Chris Thomas: There are blue collar folks building tools that are replacing costs. There are medical doctors creating tools. My buddy, he does something in healthcare with the brain, he goes, hey, look at this app I built. It's in the app store, it's on his phone. It's a camping app. He has a camping assistant and it does all the things for his camp. I'm like, dude, you're in the medical field, what are you doing building this? He goes, Claude just allowed me to do it. I was able to do it.

And then I ran into a general contractor who created an estimating tool. I was like, dude, what in the world? And he goes, well, you were talking to me about this, and I just sat with Claude Code on the desktop, and then I moved to terminal because it showed me how to move to terminal. And now I'm saving a thousand to two thousand dollars a month from a SaaS product.

I think we're getting into a new era, and I don't know what collar that era is. There's no collar era, because developers are white collar, we're supposed to be untouchable. If people are hungry enough and tired of horrible customer service, or, I want this tool to do this, they're going to go build it themselves. The more the word gets out, and the more that Bob who drives a tractor says, look, I just built this whole plot laying thing in an afternoon, that's where we're at. People come to me and say, I have an app idea. And I know they can't afford to work with me, so I go, all right, let me show you how to get started. And then they go and do it. And it blows my mind. There's no way to stop it. But if you're a SaaS platform, what are you offering that nobody else could ever build?

James Carman: With all things there's balance. The whole notion of, so and so is dead, the answer is more nuanced than that. I think there's a degree of truth to the SaaSpocalypse. Some companies just don't want to make that investment, where I have to build the thing one time and now I have to care and feed for it. It depends on how adjacent to my core business that is. Is it way out on the fringe? Then maybe I don't want to run the thing on my own, maybe I'll just pay somebody to do that.

If I have to internalize all of the things I use from a SaaS perspective day to day, run them, care and feed for them, patch them, I think people are going to quickly realize we don't want to be in that business either. But I also think I might be buying the Cadillac because I needed electric windows. Maybe I'll just build an electric window. I don't need the rest of the daggone Cadillac. I think there's going to be some erosion of the SaaS market, but I don't think it's dead. Those things are more approachable now. You can self-serve and build those one off tools that replace the things you had to buy for a hundred thousand dollars a year when you're using two thousand dollars worth of it.

Lightning round

James Carman: Now the next segment of the show is the lightning round. I don't know that I've had many other guests where I have been so excited to hear the answers. A lot of people hear lightning round and think, okay, this is just going through the motions. Well, not here at the Forward Slash. This is where things start to matter in this conversation. These answers matter. There are correct answers. There are very incorrect answers that you should not be giving. And we'll be grading it. There will be a score. We have a big algorithm. It takes a couple of months to run through everything. They can actually put out foundation models quicker than we can run the algorithm for this scoring. It's serious business, Chris.

Chris Thomas: It looks like I'm not paying attention. This is just my face. It's always confused. I have resting confused face. I'm going to win. I'm going to do it.

James Carman: What was your first celebrity crush?

Chris Thomas: My wife hates this story. I used to play harmonica and B3 organ professionally, and I used to produce music. I was in a studio with L.A. Reid, sitting on a couch, and this lovely lady sat down next to me. Her name was Alecia. I said, hey, would you like to go to Chili's? Because that's what you did in the early two thousands.

James Carman: Absolutely. If you're serious about courting someone, you take them to Chili's.

Chris Thomas: She was Chili's worthy, I will say that. And I had just gotten a gift card because it was Christmas time, so I knew it was going to clear. Alecia did not go to Chili's with me, because she said, I have to stay to record. But she said, I would love to, can you wait around? And I did not wait around. I was an idiot. Alecia is now Pink. So Pink would probably be the first one. I was tired of waiting around. I needed some ribs.

James Carman: Super Mario Brothers or Zelda?

Chris Thomas: Spicy take, I've never played Zelda, because I didn't want to be a girl in a video game. That's what I thought when I was a kid. And then Mario, I was definitely Mario. Is Zelda a girl or a boy? Can you answer me that?

James Carman: Zelda's a girl, but Link, the character you play, is a boy, from my recollection. How do you feel about Metroid? That was the big one for me when I was a kid.

Chris Thomas: No. See, I was in the Blue Ridge Mountains. Our video games were rocks rolling down a hill and being like, look how fast that went, and it made a big splash into the river.

James Carman: Metroid was the one for me where you're playing as this character, Samus, and I remember the first time they revealed that Samus was a woman. I was like, what? You've been fooling me all these years?

James Carman: When you fly on a plane, do you wear a neck pillow?

Chris Thomas: No. I have one of my own built in already.

James Carman: This one we usually reserve for around Thanksgiving time, but I want to hear your answer. How do you feel about cranberries?

Chris Thomas: You've got to get out of here. I don't even like Thanksgiving food. The best Thanksgiving we've had, we served fajitas. That's good food. And we still slept in the afternoon. Everybody took a nap. Cranberry juice, that's fine.

James Carman: Everybody's so enamored at Thanksgiving with the turkey. But when are you ever baking the turkey the rest of the year? You're not hot to trot on turkey for three hundred and sixty four days a year. Thanksgiving comes around and you say, oh my gosh, I've got to eat turkey. Okay, polka dots or stripes?

Chris Thomas: Look man, I stay away from both, because I can get easily mistaken for a picnic table. Look at that tablecloth, it's flying off. And it's really just me walking around. That's why I don't wear plaid. I used to do it a lot and then there was a picture of me with a picnic table behind me and I said, never again. Now I wear nothing but black and gray. I'm like Eeyore AI.

James Carman: Climbing a mountain or jumping out of a plane?

Chris Thomas: I feel like I'd fall real quick out of a plane. Climbing a mountain.

James Carman: I would be afraid to jump out of a plane. I would climb a mountain. Now, am I doing the thing where I'm holding on with just my hands and swinging myself over to another ledge? No, I'm not doing any of that. But if there's a nice well marked trail going up that mountain, I'm in.

Chris Thomas: My wife wants to go hiking all the time and I say no, because it feels useless. You go in somewhere and you come right back out where you went in. What was the purpose of that? So we went to North Carolina and she said, let's go hiking. I said, cool. She said, that's what you're wearing? I said, yeah. And I drove her to a mountainous golf course and said, we're going to hike in a golf cart and ride around. You're going to see the same thing, you're going to see deer, you're going to see some scenery, we're going to not be tired and miserable and arguing on the way back. That's saved our marriage quite a few times.

James Carman: My family was from Tennessee. We had hillbilly golf, so there's some uphill and downhill going on on the course. Now, we ask this question of quite a few guests. It's been a bit of a social experiment. Do you like the smell of gasoline?

Chris Thomas: I used to, but now it's different. It doesn't smell the same. I feel like I smelled gasoline everywhere in the eighties and nineties and I don't really smell it as much. Or maybe I just don't smell as well anymore. My kids love it. My son is like, I love the smell of gasoline. And I said, I was like you when I was your age. I burnt all my senses smelling gasoline.

James Carman: On a scale of one to ten, where are you putting yourself at the game of wiffle ball?

Chris Thomas: I can play some ball. At my age now I'd say a seven. If this was ten years ago I'd be a nine or ten. I'm one of those big dudes that's good at every sport. It's an anomaly that, to use your word, people are just enamored by. I was the heaviest kid in high school to dunk a basketball. But I'm actually a really good golfer. I'm on the road to being a scratch golfer.

James Carman: Here's something that really tells you a lot about a person. If the toilet paper roll is really low, but not completely out, do you go ahead and replace it, or are you leaving that for someone else?

Chris Thomas: If I see it's low, I will go and find a roll, just in case. You don't know what's about to happen.

James Carman: You could be that next person too. You're almost leaving yourself in a situation.

Chris Thomas: A hundred percent. My wife's like, there's still some on there. I'm like, nah, I don't know.

James Carman: How many redheads are you friends with?

Chris Thomas: It's a travesty, but none. My CFO once was, and then she evaporated one day. I don't know if it's a redhead thing. I was friends with a couple of redheaded folks, and it's acquired. I'm a very laid back person. Most redheaded people are not.

James Carman: You brought up Mexican food earlier. Black beans or refried beans in your burrito?

Chris Thomas: Rice. Just rice. No beans. No bean guy.

James Carman: All right, that's it for the lightning round. I feel good about this one. Again, we're not going to know for a couple of months, but I like your chances. I don't know if you won, but I'm pretty sure you didn't lose.

Where to find Chris, and the book he is trying to write

James Carman: Anything you've got coming up? Are you speaking anywhere? Do you have a book coming out? Are you doing a one man show on Broadway?

Chris Thomas: Three out of the four things you said. I'll be at Columbus AI Week in Columbus, Ohio, and then Cleveland for VC Con. And if anybody's in the south, I'll be in Tampa for Tampa AI Week. And then I think I go back to the Great Lakes for some AI conference up there as well.

As far as a book, I am actually talking to a publisher at the moment. I want to do a book, but I want it to be different. I want people to be enamored by it. The idea is it's either going to be a how to, but with stories, and it's also going to be either a coloring book, like every other page, or a pop-up book, because I haven't seen those in a while. It'd be fun. And then I'm going to focus really on audio. I'm going to read the book and hopefully kids across America will go to sleep to it. Who knows. But learning AI. I'm trying to make AI relatable.

James Carman: How do you think about translating a pop-up book to audiobook format? Is it just you're reading it and every now and then you go boo?

Chris Thomas: I hope people buy the book and then I can illustrate what just popped in their face. But the whole idea would be, if you open up the book, the process flow of the AI pipeline, that's what would pop up. That's what's different. I got early into this AI space years ago, machine learning, computer vision. I know that I'm different. And I wanted to be relatable, and it's not that hard, to be honest.

James Carman: I can say very confidently, you are different, Chris. If for nothing else but your take on cranberries. I feel like there's no one else who has as strong of an opinion about cranberries. If people want to learn about YNDR, you said it's yndr.com.

Chris Thomas: That's right. And then if they go to LinkedIn, YNDR, there's a book a call. There's also a free Claude course on there. I'm going to be putting out a lot more content. I get a lot of people asking for more content, which I've never experienced before, which is very fun. We're in the early stages of this kind of wave, and I think this is where it's good to be kind of the first to it.

James Carman: Well, Chris, it has been an absolute pleasure. I really enjoy every time we chat, and I thoroughly enjoyed this conversation. I know the audience will enjoy it. Thank you so much for coming on the Forward Slash with me today. If you'd like to get in touch with us here at the Forward Slash, drop us a line at theforwardslash@callibrity.com. See you next time.

Chris Thomas: Thanks for having me. A lot of fun.

Ship It or Skip It

Where to find Chris

Chris is on LinkedIn, and YNDR's work is at yndr.com.

Got a topic for us?