Understand This Before You Use AI To Build Software ft. Chris Guthrie

Listen Now

Chris Guthrie: Unless you actually have an understanding from a computer science level of what is happening, you are building a house of cards every single time. Bryan Fields: Chris, AI is making software so easy to build. So why can't every team just build their own internal software tools? Chris Guthrie: I mean, these are questions that I think we've both gotten very recently, and yes, it is a thing that you can do, but the difference between wisdom and knowledge is knowing whether you should do the thing that you can. I would point to the rest of the business landscape. What other industry — United States cannabis industry in 2025 was $30.5 billion, that's from Whitney Economics, right? What other industries out there are vibe coding their supply chain, ERP, customer documentation, sales processes? Where else is it happening? Not to say you can't create useful tools that hone in on what you want to do with your data, but where is it? Where's the aerospace company vibe coding their supply chain? Where's the cosmetics company vibe coding their CRM? There was that big doom-and-gloom LinkedIn cycle about 'SaaS is dead, SaaS is dead.' More and more, people are discovering there's a lot more chip than salsa right now, and I think we'll keep seeing that. So the answer is: show me another industry doing it successfully. Or do we still have that latent need cannabis entrepreneurs have to walk out on the limb with big middle fingers up, like 'look what I'm doing'? Should you do it? Bryan Fields: Do we think the idea is more of the problem — where they're like, 'Hey, my business is really lean and tight, and I see this as a cost-saving metric'? Do you think that's the origin of the idea, or is it something else? Chris Guthrie: I think there is some cost-saving, I do. But I also think there's a pathological need to be first, to be doing something nobody else is doing. Everybody's in the cannabis industry now — you're not automatically top dog anymore just by saying it. For a while you'd hesitate to say 'I'm in the cannabis industry' because people would suddenly get so interested. Now it's more like, 'Oh really? Cool, I do insurance.' So I don't really know, man. I think vibe coding tools have their place. Bryan Fields: Where I start to struggle is thinking about businesses that are vertical or semi-vertical — they've got cultivation, manufacturing, some retail, they're focused on compliance, trying to keep everything moving. Hearing some of these teams look to build tools internally just feels like another weight on the shoulder, another responsibility to figure out. It feels like a distraction — we know cultivation is solid, extraction is close, supply chain could be better, but I'm going to spend my finite resources over here instead. I wonder if that's a way to convince the brain this is a better approach because of some cost-saving metric. But I think that's the problem. Chris Guthrie: If it works, it could be a cost savings. Maybe define 'if it works' so we're all on the same page — because that's the difference, right? You could vibe code live right now, but what is 'works'? Okay — works is getting data into your system about your inventory: how that inventory is being used, when it needs to be used or consumed, where it's going to be used and consumed, and by whom, at what rate. I'm building the world's ugliest outline on my thumb here, but bare minimum, you need to be able to record that data. Then that data needs to be associated and laid out in tables in a way that's readable to the next person in the process flow, because you're not going to be in the same room anymore. That's part of why people think they can do it — they remember, 'Yeah, we ran 100 lights, it was great, we crushed it.' You're not running 100 lights anymore. You're running a thousand lights in ten different places across two buildings. So, bare minimum: can you record the information? Once it's recorded, how is it organized so others can access it? Does that information flow between departments? Does wet flower harvest translate into a drying room production scheme? Do those activities yield a new raw material input for manufacturing? It's very easy for any business — and this isn't cannabis-specific, I work in ERP for other industries too — to think their business is simpler than it is, because it all lives in someone's head. They've been doing it for 30 years, they can do it. But when you unpack the actual complexity of manufacturing, never mind cultivating a raw material for super-regulated processing and manufacturing all the way to CPG, it's much more complicated than people realize. And that risk — that all the process lives with you — is real. One thing I say to prospects, kind of a smartass line: what happens if you get hit by a bus? My business mentor told me, if you're running a business that can't operate while you take a month-long sabbatical, you're not running a business, you have a hobby. My cousin used to run rally car — you can have expensive hobbies, but that's not a business. Bryan Fields: So, yeah, back to your point about how it all lives in your head, so easy — I think that's part of the path I'd expect teams to go down. There's a lot of teams thinking, 'We could just vibe code this, Johnny's really smart, we'll hire an AI consultant.' Then you have to think: does it live in the cloud or on desktop? How do we sync information? How do we handle unit transfers — if someone orders 10 versus 25, do they know if units differ? If we're doing calculations or adjustments, or running things uniquely, which most teams do out of the box, you have to adjust for all of that. And my favorite part about software is that editing one field can have corresponding effects on other fields down the road that you didn't realize up front. AI, as powerful as it is, can't infer that you're changing something on your cultivation page that should also account for finished goods down the line at a certain rate. It doesn't know that. So you set yourself up for a disaster path — easy to get started, but so hard to take from an MVP, 'is it working,' to something functional you can actually deploy as a company. Chris Guthrie: Most of the cannabis operators I work with lovingly call their employees idiots. Is what you're vibe coding on Saturday and Sunday going to withstand the full-contact violence of your production and cultivation team? They're busy, they're sticky — what kind of UI did you build for them? Are there stopgaps to prevent missing data points? And going back to vibe coding — unless you have a git repository and a development environment in front of you, Claude is doing all that code and you don't know what's going on back there. So worst case, you have something that works for you and a couple of associates for a couple of months, but then— Bryan Fields: Yes, a couple of months — let's make it even worse. Chris Guthrie: And then two months down the road, getting ready for production, you discover a crucial field doesn't map, a workflow doesn't tie, something is broken. You cannot find it, because that was a vibe code session from a month and a half ago, and who knows if that decision tree is still left in your model. You will not find that mistake. You can review the code all you want, but this is what's actually wrong with cannabis industry technology specifically: tech debt. You will incur tech debt vibe coding. Anybody who's spent a weekend building an app knows you get to something almost good, then at the last minute it breaks, you try to fix it, and it's a cascade of failure. Vibe coding makes you feel awesome, and ChatGPT and Claude are there to stroke your ego — but unless you actually have a computer science-level understanding of what's happening, you are building a house of cards every single time. It will not stand any sort of violence. Bryan Fields: It does feel so good, and I think that's part of the problem. It's addictive, like a video game — you have a problem, you get started so quickly, Claude shows you a working prototype in 15 minutes, and you think, 'I'm going to solve my problem in 10 hours, roll out an enterprise-grade tool, go back to my day job, and be a hero.' Teams forget that other businesses have spent enormous amounts of money learning how to set up correct infrastructure — data protection policy, privacy policy, computer infrastructure policy — to work through the surprising edge cases, so teams can do what they do best: sell the molecule, sell THC. Bryan Fields: Right, and from a software ecosystem point of view, software is about setting something up, running it, seeing what happens, and correcting mistakes. The cannabis software industry has been pushing those dominoes down for maybe 12 years, tops. Chris Guthrie: Right. I work with a Microsoft platform called Business Central that we've customized for the cannabis industry — we're all from the industry ourselves. We have the benefit of millions, tens of millions of hours of use cases, business scenarios, breaks, new business processes, accrual models, item structure, master data, numerical sequences. You're not going to ask Business Central to do anything it hasn't been asked before. Without naming names, you will constantly ask cannabis-specific software to accommodate an item structure, a BOM configuration, multi-output finished goods from multi-input production schemes — and if the developers haven't seen it before, the computer only knows what it knows, and it won't solve it. That's why cannabis operators keep running into walls: 'it's coming in Q3, it's coming in Q4' — never mind the changes the metric middlemen ask people to accommodate. You can only solve for problems you've encountered. Bryan Fields: Just to be clear — Microsoft, as in the Microsoft, the Seattle-based, Bill Gates-founded Microsoft? Chris Guthrie: We have to give some of these companies their flowers — they've been around a long time, they've established infrastructure and brand credibility you can trust. Setting up software environments involves a foundational level, taking an idea to fruition through steps and environments, so you don't just ship it and hope it works. It has to go through testing so it doesn't break other things or reset your whole module — because the last thing a strapped team wants is a step where clicking a button doesn't work, and you're stuck waiting two weeks for a fix, or 'Elliot's on PTO for his wedding, he's the only one who knows the code, I can't fix the spreadsheet ERP.' That's an infrastructure and code-integrity issue. Going back to whether you can vibe code — Microsoft spends most of its time maintaining stability, repeatability, and deployability of its code. Business Central now has a standard set of fields and APIs that doesn't change — you don't customize your ERP core anymore, that's an industry-wide shift. The core computing structure is hardened and set, and you choose how to interact with it. That's what makes Business Central and 365 Cannabis powerful — it's stock cultivation and manufacturing but malleable enough to accept nuance, because every grower I've ever met was sure they were the only person to ever grow cannabis successfully the right way. We can accommodate all that because there's solid infrastructure backed by Microsoft. Love or hate big tech, they've got a battleship, and you can build a house on it and they'll sail you around. Bryan Fields: Do you think people sometimes misrepresent the cost involved? Getting started is easy because Claude makes it so accessible, but as you rack up API bills when things don't work correctly, your $10,000 budget isn't sufficient. Plus hosting costs are expensive — you want multiple people using it at different sites, so it has to be in the cloud. Chris Guthrie: Microsoft and Amazon do a good job of saying, 'Happy to host your project, but it's going to cost $1,500 a month,' and you're like, 'For what?' It's, 'Welcome, I've got this box, you want to put your stuff in it? That's what it costs.' People think, 'I'm going to save $3,000 a month' — maybe, or you could spend $4,000 a month on a non-working product that drives your team crazy, and you're wondering what you even did, wishing you had that $50,000 back. Bryan Fields: Yeah, or invested that $50,000 in a system that comes with hosting and is stable. Chris Guthrie: If something seems free, you're the product. Bryan Fields: Right, and that's exactly the promise of it. Chris Guthrie: You make huge strides in that first 15 minutes, but every hour after, you move the needle less. God forbid you build something into your vibe-coded ERP that requires AI tokens — not only did you spend tokens vibe coding it, now it consumes tokens every time you run a query, a rendering, or a relational ask. What's that going to do? Bryan Fields: That's the part I think people want — they've got all this data and want predictive insights from it. But that's where vibe coding your own ERP leads to more problems, because without directional access to clean, accurate data, you're just handing Claude a pile of information and saying, 'Help me understand this,' and it's, 'Good luck.' Chris Guthrie: Claude will say, 'Wow, great data set, really insightful, great job — I can't do anything with this. This is a pepperoni pizza, correct?' And it's like, 'No, Claude, go home, you're drunk.' But $50 back in the coin slot has just opened up on your laptop — please deposit $75,000. Bryan Fields: What you said about needing to get predictive is more and more what I'm hearing from organizations: 'We've been going along, but I need to look into the future a little, I need to be ready. We've survived by the skin of our teeth, by sheer determination.' Chris Guthrie: What ERP is supposed to do is use your data to make predictive decisions — know what raw materials you need, when you'll need them, labor demands, predicted inventory levels and costs. That's the actual information you need. So are you a computer scientist? Great, you told Claude your business flow and it told you what you need. But can you turn descriptive data into predictive tables — if this number, then this number, then this number, that's a yield projection? The cannabis industry is built on a descriptive data capture structure. All seed-to-sale software started as a Metrc skin or a BioTrack skin. People tried to connect with Leaf Data, but that door just hit a brick wall. People need ERP for predictive, but I don't think the data in cannabis tools today even allows predictive, because it's just a record of what happened. You'll still have to ask AI to extrapolate the future. Are you pathologically trained by the cannabis industry to reinvent the wheel and be first with a brand-new idea, or are you willing to accept humanity has already done this work for you? ERP is the wheels of business right now. Blend a foundational, trustworthy ERP with a data scheme that lets you pull the right variables, and you can get predictive insights — if you have that foundation. Without it, you're saying 'this guess times this guess plus this Claude guess gives me this guess.' And then I go back to: is this the best use of your team's time? If capital is scarce, are we spending money on what we do better than anyone else, or distracting ourselves from our core focus — cultivation, retail, extraction? If a cannabis company asks 'what do we do better than everybody else' and the answer is 'vibe coding,' turn off your lights. That's not what your team does best. Bryan Fields: Sure, but I think what happens is they think, 'We could save cost, we've got two smart employees, we'll hire an AI consultant' — a very popular term I've heard lately. I have no idea how that turns out, except thinking it might be lighting money on fire, and in six months deciding it was a massive waste of time, energy, and money. Maybe start with the goal in mind — because if it's saving money, it's probably not going to save you money long-term. Chris Guthrie: No, because you make huge strides that first Saturday and Sunday, but the quality and flow of every iteration after keeps shrinking. There's no way to keep up with that diminishing return building an AI tool yourself. An 'AI consultant' — what is that job, really? Chris Guthrie: One thing I heard in Chicago — Chris Vis of Blaze, on a panel we were both on, said it really well, and it's something I admire about Blaze: they've got very strong data governance rules. All their base information and instructions are stored in a vault, a repository, because they learned early that Claude and GPT drift mid-session — it stops looking at your original data set and starts incorporating something you didn't ask for. So first — do you know how to use AI? Do you know if a problem is even solvable by AI? People think they do, that's why they think they can build a vibe-coded ERP or CRM. But it's like a non-cannabis person walking in and saying 'I'm going to start a cannabis company' to someone who was a backpack boy in 2009 — 'Yeah, I've got an Arcata Trainwreck cut my mom's been saving for me, we're going to build a business.' That sounds silly. It's the same insult when people say 'I'm going to build an ERP over the weekend with Claude' to computer scientists and ERP practitioners. You don't know what you don't know — those are the holes you fall into. I don't want to trash AI building, it's been a fun project of mine — I've got 13 Claude projects right here. Bryan Fields: It has to fit into the chain. From conversations I've been hearing, people misallocate how that thought process works. What's the goal? If it's saving money, probably won't save money. If it's improving data structure for predictive decisions, AI can help — but that requires understanding how the tools work and having a realistic view of what they can and can't do. Chris Guthrie: Right, because AI isn't creative. If you ask it to create something, it has the whole internet, the collected knowledge of humanity, to pull from. Are you sure it's looking at the right context, data, history? That's what Chris is doing at Blaze — every query is pointed at Blaze's own data. So do you know how to corral your queries? Are you sure you're asking AI a question it can actually solve? Are you willing to rest your $5 million business on a piece of software that's a month old? That's bananas to me. Bryan Fields: As the industry evolves and big players evaluate a potential partner's tech stack, and they hear 'we built our own CRM, we vibe-coded this dashboard,' they think, 'I'm petrified to see what your books look like.' There are certain boxes businesses want checked to know you're the right partner. The industry was so entrepreneurially focused — 'we have to do this ourselves' — but for some things you need the tools that are already established, you just need to find the right partners. Chris Guthrie: The only thing you needed in early cannabis to make millions was the balls to do it. Bryan Fields: What do you think they need now? Chris Guthrie: You still need balls, but also caution and patience. Grow these businesses like a plant — you can't put it in too big a pot right away or you'll stunt it, you can't overfeed it or over-borrow. There was a panel in Chicago at Ignite about how M&A this year and next is going to be fast and furious. A friend told me the pitch now for ERP is: if you want the companies doing M&A diligence to take you seriously, you better show clean books, clean processes, a full list of liabilities and assets with depreciation, and records of how the business has performed. Otherwise, why would they buy you? Is it a shiny spaceship someone paid $32 million for seven years ago that's now worth ten cents on the dollar? Bryan Fields: Most of those companies are using Google Sheets for things outside their core use. Chris Guthrie: That shocks me most — heavy Google Sheets use for things it's not built for, and instead of moving to purpose-built tools, teams skip straight to vibe coding, digging themselves another hole. Bryan, I don't know a gentler way to say it: if a company goes from Google Sheets straight to vibe coding, watch out — that's a path to failure. Please don't do that. Bryan Fields: I think this is where the industry is starting to separate. The big players with means know what they do well and don't, and they'll offload what they don't do well. Other players doing fine in one market think, 'How do we grow?' and default to cutting costs by building tools internally. I can't shake these companies enough — think about what you're investing in, because the approach needs to change. It doesn't have to be about cutting cost, it can be about optimizing process — understanding where your costs go, what drives them, and where you can save. Chris Guthrie: Right, and you're not going to vibe code a true, reliable standard cost. I've got some Josh D. Motorbreath right here — these guys know their cost. Do you know precisely what went into that flower, the jar, the label, the wait time? Seed-to-sale doesn't do that. By the time you finish your Saturday vibe coding, you might think you've accounted for it all, but you're pushing it down the hill. And what if it doesn't work? Bryan Fields: What if it doesn't work? You're a CFO — take your annual salary, divide by days and hours worked. That whole weekend you probably spent $7,000 of company time vibe coding an ERP that doesn't work, and now you'll spend another $7,000 to go back. Just in labor cost, you can measure how much your vibe coding expedition cost — keep track of it. Bryan Fields: Do we think this is part of the origin issue — companies had patchwork software that didn't connect well, got disappointed, and now think 'I could just build my own'? Chris Guthrie: Change management is the number one reason ERP projects fail to get adopted. I was in an 18-month cycle with a top-five company in the country, and the overall feeling was, 'This is what we need to do, but we're in 14 to 16 states, multiple facilities, a lot of it from acquisitions' — not the same culture, processes, or data structure. Aligning hundreds of people to define one national process, deploy it, and train on it is a massive lift. Once you get into the interdependencies of everyone in an organization, it's complex, and fear of not having the bandwidth to organize the business keeps people from doing it. But the house of cards eventually falls. You can ignore the elephant in the room for a while, but eventually you have to deal with it. Bryan Fields: There's never a good time — there's always something going on, always a problem. This needs to be a priority that happens yesterday, and that's the fear I have for companies that avoid it because change is hard. Chris Guthrie: And that's part of the problem. But if not now, then when? The best time to plant a tree was 20 years ago, the next best time is today. Bryan Fields: Everybody's just been trying to survive the past two years. Nobody was thinking about a 5-10-15 year plan in 2023 — it was nothing but broken dreams. Chris Guthrie: Sure, but that's the frustration — teams thinking about vibe coding to cut cost could use that same resource to structure everything onto the same page, so the same teams tracking things in Ohio and Virginia are tracking the same elements. Bryan Fields: Yeah, and maybe it's Stockholm syndrome from being in the cannabis industry too long — building your business out on another shaky branch. How do you get them to change, knowing it's a massive uphill battle? Chris Guthrie: First, I make sure I understand what they're actually asking for. I've had discovery calls where someone says, 'I need some software to fix my business, man,' and I reflexively say, 'I don't sell that — I will help you measure your business.' Going through an ERP exercise is the best, most comprehensive business analysis and process-improvement activity there is, because you have to get everything into the system for it to function. I'll show them everything and lay out the project, and they say, 'I can't afford that.' My next question is: what happens if you do nothing? If you do ERP, you'll know these things. If you don't, what will happen? Some say 'I'll figure it out,' and I know they're off on the Oregon Trail about to die of dysentery. Bryan Fields: Good luck to you. Chris Guthrie: If instead it's, 'You've given us a lot to think about,' there's hope. The only reliable way I've found to change minds is to give them the information and hope to hear back in six or eight months: 'We tried this, it didn't work, we know we need ERP now.' Done correctly, ERP ROI is two years minimum — from not missing orders, ordering on time, being able to predict your business rather than just looking at descriptive data, or a Google Sheet where it's, 'Look at all my numbers, isn't it beautiful?' Bryan Fields: Look how many colors I've got on my tabs, I spent three hours color-coding this. Chris Guthrie: Oh, do you? Great, that's a $1,200 color scheme. Bryan Fields: I was about to shred someone, but— Chris Guthrie: No, no, we keep names out — caught myself. Bryan Fields: I have a real problem with Google Sheets, and the colors are triggering — you look at it and it gives you anxiety. I ask what it means, they say this number's good, this number's bad, and when I ask how they got those numbers, they say, 'I don't know.' We got green and red, but we'll never know the difference. Chris Guthrie: And I know they spent real time considering those colors — 30% orange for this, 60% for that — using brain power, time, and money literally on fire. Bryan Fields: Are there any takeaways for people on the fence — not just ERP-specific, but given everything you've seen in software tools and teams — that people aren't considering heavily enough about how fast the world is changing? Chris Guthrie: First, the last episode you did, I listened to it twice, it blew my mind, such a good conversation — because I think a lot about Schedule 3, and what happens if the Epstein files blow up and Trump has to decriminalize or legalize cannabis just to shift the conversation. If the FDA gets involved, there are laws, procedures, standards, and inspections that will instantly govern cannabis, similar to tobacco or vape — cosmetic, food-grade standards. There is no Google spreadsheet in the world that meets those standards. The FDA measures by traceability and defensibility — that's how that platform works, and pointing to a Google Sheet will be triggering. Bryan Fields: And I'd say more so 'when' — I think that's the better way for teams to think about it, so they're prepared instead of caught behind. It feels like we're always very reactive. Chris Guthrie: Yes, and staying reactive is the same as depending on descriptive data. I work with a lot of Canadians, so I know the phrase — you've got to skate to where the puck is going. The FDA will get involved; when I say 'if,' it's just because the process is messy, and executive orders don't work like snapping your fingers — it'll take time to get clarification on what Schedule 3 means. But the benefit is the FDA governs by rules that are already published, so the bones of what the rules will be already exist. Shops need to stop telling themselves and their shareholders they don't know what's going to happen — you can predict this. You'd better be able to report what's in that jar the same way Philip Morris reports what's in a pack of Marlboros, Oil of Olay reports what's in a jar, or Hostess reports what's in a cupcake. It better have a change log, be publicly hosted, follow GAAP accounting principles, and follow GACP — Good Agricultural and Collection Practices — for cultivation. These rules are out there, you can know this. The question is whether you'll set your business up for the next phase of the industry, because it's not balls anymore — it's careful, thoughtful prediction based on your current data set. That's how business works. Bryan Fields: What question do you wish more people asked you? Chris Guthrie: What is the actual ROI of ERP — is it really two years? It depends on the size of the project; when I say two years, I'm looking at something like a $500,000 project. Bryan Fields: Is two years too long for some teams? To me that's fast, but I imagine some teams say, 'No way, Chris, two years, we'll never make it.' Chris Guthrie: The one phrase I never want to hear again is 'one year in the cannabis industry is like one year in any other industry.' Bryan Fields: Sure, that's great, but I think that's slowing down. Chris Guthrie: You've got to know where the ROI is. My bullseye goal is to save an organization an hour a day across ten people — not walking back and forth, not harassing someone for their cultivation spreadsheet, not wandering the warehouse wondering if the soil came in. Ten hours across ten people a day — do the math. Save your accounting people even 30 minutes a day, that's two and a half hours a day across working days — instant, measurable ROI. Never mind the better decisions you make with real strain metrics, or knowing a job predicted to cost X actually cost $500 more once you record actuals — where's that leak? Or it cost $600 less — you don't know unless you measure it in real time, organize it, and then ask AI to find the patterns: compare actuals versus forecast costs for this round of distillate or batch of flower over four months. That's what AI is for — that's what you can vibe code around: show me the patterns in the data. Bryan Fields: It's triggering me, Chris. So Chris, for our listeners who want to get in touch and learn more, where can they find you? Chris Guthrie: No, I'm not taking that bait. Bryan Fields: No man, I laid it up. Chris Guthrie: Okay — I'm Chris Guthrie, I work at 365 Cannabis, about six years now. We deploy Business Central and cannabis-specific extensions across Canada, the United States, Europe, and Australia. You can find me at bigfan206 on Instagram — I'm not on there a lot, but I still lurk. You can email me at cguthrie@365cannabis.com. Bryan Fields: Thanks for taking the time, this was a lot of fun, dude, I so appreciate it. Chris Guthrie: Thank you for having me on for the season closer.