Into the Gnar

Replit just made OpenAI the default behind its coding agent instead of Claude, and the debate over which AI company to stand behind keeps getting louder. But switching LLM providers is easier than switching most SaaS products. The hard part comes before the model: documented processes, clear KPIs, and a source of truth everyone trusts. Most companies skip that step and go straight to flipping the AI switch. So what should a CEO actually do before choosing a model? That’s what this episode is about.   In episode 14 of Into the Gnar, it’s just us, Mike Stone and Nick Maloney, co-founders of The Gnar Company. We sell AI transformation to mid-size B2B companies, and we’re right in the middle of our own. We also just made Anthropic’s first Select Tier of partners. Getting in took two reviewed production deployments, a public case study, and 10 of us passing the Claude Certified Architect exam (nobody got a perfect score, and all of us had to study).   We’ve learned the hard way too. We built so many one-off internal apps that each did 5 or 10 percent of what we needed, so now we map friction across every department before anyone writes code. Nick compares AI to a bigger piece of heavy equipment: it moves more dirt, but only if someone trained is in the seat. And being honest about it costs us deals, because a lot of executives want ROI next quarter, and about 80% of AI strategy work isn’t AI at all.   We got into:   Replit making OpenAI its default, enterprise vibe coding, and why “vibe coded apps are the spreadsheets of 2026” Apps that are “cheap to build but expensive to operate,” and the security risk of back-office tools with no controls Gemini 4 Argon’s pricing, a price that doubles in two weeks, and whether a cheaper Fable would change how we work The last mile of AI ROI, why cutting headcount isn’t the play, and how the bottleneck just moves (for us, to code review) Getting AI to stop writing walls of text, with tricks borrowed from military radio communication “Automating bad processes is not a good thing to do,” and why SOPs and KPIs come before any model Small, isolated agents, mixing models by task, and self-hosted LLMs that could match Opus in 6 to 12 months   If your company is debating models before it has documented how work gets done, start here.   Take our free two-minute AI readiness assessment at ai-assessment.thegnar.com    New episodes every other Monday.   TIMESTAMPS: 00:00 Switching AI Models 02:09 AI ROI and Bottlenecks 08:09 Fixing Verbose AI Output 11:45 Replit Switches to OpenAI 14:54 Vibe Coding Security Risks 21:06 Gemini 4 Argon Pricing 26:21 Anthropic Select Tier Partner 33:09 Processes Before AI Models 40:31 Modular AI Agents 43:29 Self-Hosted LLMs 46:45 AI Readiness Assessment   Hosts:    Mike Stone, Co-Founder of The Gnar Company https://www.linkedin.com/in/michaelvstone/    Nicholas Maloney, Co-Founder of The Gnar Company  https://www.linkedin.com/in/nicholasmaloney/  Links Free two-minute AI readiness assessment: ⁠https://ai-assessment.thegnar.com⁠  The Gnar Company: ⁠https://thegnar.com⁠  If you've run a software project that went sideways, or one that finally went right, drop it in the comments. We read all of them.   Watch/listen/Subscribe:  Spotify: https://open.spotify.com/show/0JoqOMNJf8szkisHWebLwg?si=03a6d10a765146ef Apple Podcasts: https://podcasts.apple.com/au/podcast/into-the-gnar/id1896780159  Youtube: https://www.youtube.com/@thegnarcompany   Subscribe so the next episode finds you.   Produced in partnership with OBEY Creative, building podcast-driven content engines for B2B founders. Learn more at⁠ ⁠⁠https://obeycreative.com⁠

Not sure if your team is actually ready for AI? We built a free 2-minute assessment that gives you a straight answer → ai-assessment.thegnar.com/  Your company rolled out AI. The answers still come back wrong, generic, or contradicting your own documents. The problem probably isn't the AI.   Samantha Maxim runs the AI strategy practice at The Gnar. She's spent her career in product roles, including running a learning platform portfolio at Boston Consulting Group (and she landed in QA by accident at the very start). She's also the first person from inside The Gnar we've brought on as a guest.   Her take: "You can't just automate a bad process, and you can't just automate something that isn't working." AI only works as well as the information behind it. At most companies, that information is spread across drives, inboxes, and a few people's heads.   Samantha breaks down information architecture in plain language: where your information lives, how it's organized, and which source wins when two of them disagree. She shares the worst shared drive she's ever opened, at a 150-person company with unencrypted database backups sitting next to weekly meeting notes. She walks through how she tells Claude which systems to trust first, and why a comment from a meeting shouldn't beat your requirements doc.   She also gives mid-size companies a way to start without a year-long cleanup. Set a document retention policy. Move anything older than 10 years into a "to be deleted" bucket. Make each department review the files it owns.   Plus, the news: ChatGPT shows up inside Microsoft Word, and an OpenAI agent gets unauthorized access to an Australian government portal. Samantha has a lot of feelings about that second one.   In this episode, we cover: What ChatGPT inside Microsoft Word means for your file permissions and access controls Why the "one AI app for everything" future looks more like streaming than cable What an OpenAI agent getting into Australia's Medicare portal teaches anyone building agents Why your AI gives bad answers (usually the prompt or the context window) Information architecture, explained for people who don't design websites How to set a source of truth your AI tools can trust Quick wins for cleaning up company data without a big-bang project Why turning on Copilot across the company is a form of AI theater If your company has the AI tools but your data still lives in five places and three people's heads, start here.   Take our free two-minute AI readiness assessment at ai-assessment.thegnar.com.   New episodes every other Monday.   TIMESTAMPS: 00:00 Why We Hired Out the Lawn 03:03 The Worst Shared Drive Ever 05:16 Meet Samantha Maxim 06:12 What an AI Strategy Lead Does 07:01 You Can't Automate a Bad Process 11:46 Why AI Gives Bad Answers 14:10 LLMs Don't Actually Think 19:31 ChatGPT Comes to Microsoft Word 23:36 Will One AI App Replace Them All? 27:31 OpenAI Agent Breaches Medicare Portal 30:55 Who's Responsible When AI Goes Rogue 32:09 How to Test AI Agents 33:40 AI Agent Permissions and Guardrails 37:08 Why AI Regulation Is Coming 39:24 What Is Information Architecture? 44:59 How to Set a Source of Truth 49:00 Quick Wins for Messy Company Data 54:18 MCPs and Choosing the Right AI Tools 01:00:10 Rapid Fire Hosts:    Mike Stone, Co-Founder of The Gnar Company https://www.linkedin.com/in/michaelvstone/    Nicholas Maloney, Co-Founder of The Gnar Company  https://www.linkedin.com/in/nicholasmaloney/    Guest: Sam Maxim, AI Strategy Manager at The Gnar Company https://www.linkedin.com/in/samanthabmaxim/  Links Free two-minute AI readiness assessment: ⁠https://ai-assessment.thegnar.com⁠    The Gnar Company: ⁠https://thegnar.com⁠    If you've run a software project that went sideways, or one that finally went right, drop it in the comments. We read all of them.   Watch/listen/Subscribe:  Spotify: https://open.spotify.com/show/0JoqOMNJf8szkisHWebLwg?si=03a6d10a765146ef Apple Podcasts: https://podcasts.apple.com/au/podcast/into-the-gnar/id1896780159  Youtube: https://www.youtube.com/@thegnarcompany   Subscribe so the next episode finds you.   Produced in partnership with OBEY Creative, building podcast-driven content engines for B2B founders. Learn more at⁠ ⁠⁠https://obeycreative.com⁠

Not sure if your team is actually ready for AI? We built a free 2-minute assessment that gives you a straight answer → ai-assessment.thegnar.com/    If you've bought the AI tools, rolled them out, and still can't say what actually got better, this one is for you.   Brian Felker spent 15 years selling enterprise technology and watched most of it get bought and never really used. Now he runs The Growth Nerd, an advisory firm out of Nashville that goes into mid-market companies and fixes how the work actually moves. Two weeks before this recording, he published a book on exactly that problem: Operational Modernization: Closing the Continuity Gap.   His line on it stuck with us: "AI can automate work. Automation and improvement are not the same thing." Automate a broken process and all you've built is a faster broken process.   Brian walks through the three ideas at the center of his book. The continuity gap: technology moves continuously, but companies only transform once every couple of years. Human middleware: people doing manual work that software should be doing for them. Operational debt: what piles up in the gap between those two. He tells the story of a commercial real estate insurance company drowning in spreadsheets and email chains, and how consolidating their data saved up to 10 hours a week per person. He also names "AI theater" directly, the seats you bought that nobody's actually using for anything that moves the business.   Brian and Mike get into who inside a company should own this work, and why Brian thinks the COO's job and the operational modernization job aren't the same one.   Find Brian at The Growth Nerd and pick up his book, Operational Modernization: Closing the Continuity Gap.   In this episode, we cover: Why the frontier AI models converged on the exact same price within two days of each other What Apple's near-billion-dollar bet on renting Google's models says about build versus buy The continuity gap, human middleware, and operational debt, defined in plain language A real client story: how one company saved 10 hours a week per person by fixing how their data moved The most common form of AI theater companies mistake for real transformation Why nobody currently owns operational modernization inside most companies, and who should Why the economics of continuous improvement finally work now, when they never did before If you've told your board you're leaning into AI and can't point to what's actually improved, this is the episode. New episodes every other Monday.   TIMESTAMPS:  00:00 Why AI Automation Fails 01:23 Meet Brian Felker 04:21 AI Models Become a Commodity 10:33 Apple Rents AI From Google 15:36 The Continuity Gap 21:06 Why Improvement Is Cheap Now 26:56 Where AI Actually Works 34:05 Spotting AI Theater 40:25 Who Owns AI at Your Company 43:48 Rapid Fire   Hosts:  Mike Stone, Co-Founder of The Gnar Company https://www.linkedin.com/in/michaelvstone/    Nicholas Maloney, Co-Founder of The Gnar Company  https://www.linkedin.com/in/nicholasmaloney/    Guest: Brian Felker,Founder & Managing Partner of The Growth Nerd https://www.linkedin.com/in/brianwfelker/ https://www.thegrowthnerd.com/ Pick up his book, Operational Modernization: Closing the Continuity Gap.   Links Free two-minute AI readiness assessment: ⁠https://ai-assessment.thegnar.com⁠  The Gnar Company: ⁠https://thegnar.com⁠  If you've run a software project that went sideways, or one that finally went right, drop it in the comments. We read all of them.   Watch/listen/Subscribe:  Spotify: https://open.spotify.com/show/0JoqOMNJf8szkisHWebLwg?si=03a6d10a765146ef Apple Podcasts: https://podcasts.apple.com/au/podcast/into-the-gnar/id1896780159  Youtube: https://www.youtube.com/@thegnarcompany Subscribe so the next episode finds you. Produced in partnership with OBEY Creative, building podcast-driven content engines for B2B founders. Learn more at⁠ ⁠⁠https://obeycreative.com⁠  

Not sure if your team is actually ready for AI? We built a free 2-minute assessment that gives you a straight answer → ai-assessment.thegnar.com/  If you've told your board "we're leaning into AI" and couldn't say what any of it is actually costing you.   Matt Crane runs MGMT Boston, a media company built around covering venture-backed startups across Greater Boston, more than 1,600 pieces of coverage and counting. He also runs a private community of almost 950 operators. Which means he hears two versions of every AI story: the one companies post about, and the one they admit over coffee.   Those two versions rarely match.   Ramp's data on more than 70,000 businesses shows Anthropic and OpenAI running neck and neck for corporate card spend, flipping back and forth every time either one ships a new model. 56% of businesses now pay for some AI product. That leaves 44% paying for none of it, even though usage inside those companies is almost certainly higher than the invoices show. Copilot rides along in the Microsoft license and nobody sees a line item.   Matt's heard the other side of that curve too. One company, he says, blew through $500 million in tokens in six months and torched its budget for the year.   That's the second wow. The one where finance sees the bill.   We also get into why every new AI product now gets measured against a single question ("can't I just do this in Claude?"), why Klarna's attempt to build its own CRM in-house didn't survive contact with reality while Salesforce made itself agentic and got paid for it, and why Matt still writes every single event invite by hand.   Then we flip the mic. Matt spends the back half of the episode interviewing us, and we don't hide from it: how The Gnar Company rebuilt its client base, its pricing, and its own AI stack around this exact gap. We're still building the reporting to track token spend at the project level. Ask us again in six months.   Also settled, definitively: orange is the best Starburst.   Find Matt at MGMT Boston and on his show Code and Cobblestones with Will Lehmann at Step Function Ventures. In this episode, we cover: Why 44% of businesses still don't pay for any AI product, and why the real usage number is higher What Ramp's data on 70,000+ businesses says about actual AI vendor loyalty Why Claude has become the bar every new AI product gets measured against What Klarna's reversal and Salesforce's headless bet say about which software survives AI Why Matt still writes every event invite by hand, no exceptions How MGMT runs 1,600+ interviews and a 950-person community with a tiny team What The Gnar Company changed about its clients, pricing, and stack this year What a CEO should actually ask for before their next AI renewal If you want to know what your peers are really doing with AI instead of what they're posting about it, this is the episode. New episodes every other Monday.   TIMESTAMPS: 00:00 Who Is Matt Crane 02:04 Wall Street To Startups 05:50 Building MGMT Boston 06:37 AI Hearing Glasses Raise 09:37 Boston Startup Trends 12:53 Claude Vs OpenAI Spend 18:26 AI Renewal Red Flags 19:48 MGMT's Interview Flywheel 24:12 Why Events Burn Out 25:47 How CEOs Buy Software 27:31 Vaporware Vs Claude 28:26 Claude Raises The Bar 29:22 Salesforce Vs Klarna AI 31:56 Lovable Plus Claude Stack 35:21 AI Dead Ends 36:58 Human In The Loop 38:14 Why Matt Writes By Hand 40:00 Matt Interviews The Hosts 40:39 Gnar's Big Pivot 42:54 Gnar's Claude Token Spend 46:54 Gnar's Next Chapter 50:54 Rapid Fire Questions 53:05 Free AI Assessment Hosts:    Mike Stone, Co-Founder of The Gnar Company https://www.linkedin.com/in/michaelvstone/    Nicholas Maloney, Co-Founder of The Gnar Company  https://www.linkedin.com/in/nicholasmaloney/    Guest: Matt Crane, Founder of MGMT Boston https://mgmtboston.com/  https://www.linkedin.com/in/matthewhcrane/ Links Free two-minute AI readiness assessment: ⁠https://ai-assessment.thegnar.com⁠  The Gnar Company: ⁠https://thegnar.com⁠  If you've run a software project that went sideways, or one that finally went right, drop it in the comments. We read all of them. Watch/listen/Subscribe:  Spotify:https://open.spotify.com/show/0JoqOMNJf8szkisHWebLwg?si=03a6d10a765146ef Apple Podcasts: https://podcasts.apple.com/au/podcast/into-the-gnar/id1896780159  Youtube: https://www.youtube.com/@thegnarcompany Subscribe so the next episode finds you. Produced in partnership with OBEY Creative, building podcast-driven content engines for B2B founders. Learn more at⁠ ⁠⁠https://obeycreative.com⁠

Not sure if your team is actually ready for AI? We built a free 2-minute assessment that gives you a straight answer → ai-assessment.thegnar.com/  If you've bought AI in the last eighteen months and couldn't tell your CFO what it's costing you per month, this one's for you. We've felt this gap ourselves. You know the shape of it. A Copilot license here, a personal ChatGPT account on the company card there, a vendor who just raised your renewal because they bolted on an AI feature nobody asked for. The bill keeps climbing. Nobody can point to the line that's causing it. This week, we go solo and go straight at the number almost nobody in our audience has: what AI is actually costing our business, and why almost nobody can answer that with a straight face. Not the vendor pitch version. The invoice version. A study of 150 business leaders found that not one of them reported a cost reduction from AI. Two thirds said costs went up. Sixty percent are running seven or more separate AI platforms with no one person managing any of it. And by one economist's math, the companies selling AI models and apps are running negative margins funded by investors, not customers, while the hardware makers are pocketing over forty percent. We also get into two stories that landed the same week: insurance carriers quietly writing AI liability out of standard business policies, and private equity circling old-school "systems of record" like Workday, because the boring backend software turns out to be the one thing AI can't easily replace. The main point is simple: you can't manage what you never measured. Most companies didn't instrument anything before they started spending, so now the bill shows up as a surprise instead of a line item. Find the leaks. Time one job, before and after. Match the model to the task instead of running a frontier model on something a cheap one could handle. Name one owner. In this episode, we cover: Why nobody can tell you what AI is actually costing their business How subsidized subscriptions are hiding your real per-project spend Why insurance carriers are quietly excluding AI from standard coverage What Silver Lake's Workday bid says about which software actually survives AI Why "systems of record" are becoming more valuable, not less How to tell a real AI cost increase from a vendor's price hike in disguise Why matching model size to task size is the cheapest fix most companies skip How to build an AI spend baseline in one afternoon, no platform required If you've been buying AI on faith, this is the week to go find the invoices. New episodes every other Monday.   TIMESTAMPS: 00:00 The AI Spend Problem 01:49 Claude Cost Telemetry 03:27 Subscription vs Token Pricing 08:51 Stripe Acquires OpenRouter 11:58 Frank and the 80/20 Trap 16:06 Building Around Your ERP 19:12 Insurance Drops AI Coverage 25:46 Workday Buyout Explained 31:03 AI Costs More Than It Saves 39:21 Why AI Spend Is Invisible 41:40 No Telemetry? Start Here 44:10 Build vs Buy AI Tools 47:23 Key Takeaways and Next Steps Links Free two-minute AI readiness assessment: ⁠https://ai-assessment.thegnar.com⁠  The Gnar Company: ⁠https://thegnar.com⁠  If you've run a software project that went sideways, or one that finally went right, drop it in the comments. We read all of them. Subscribe so the next episode finds you. Produced in partnership with OBEY Creative, building podcast-driven content engines for B2B founders. Learn more at⁠ ⁠⁠https://obeycreative.com⁠