23: Your AI Stack, Your Rules: Local Models, Personal CRMs, and the End of the Token Subsidy

23: Your AI Stack, Your Rules: Local Models, Personal CRMs, and the End of the Token Subsidy

Most people are still renting their AI workflows. Paying monthly for tools that go stale, burning through tokens on autopilot, and building on top of platforms that could triple their prices overnight. This week, Sam and Will get into what a smarter, more sustainable setup actually looks like -- and why the people thinking carefully about this now are going to be in a very different position when the economics of AI eventually reset. Sam breaks down the personal assistant he's built using Claude Code connected to a private GitHub repository -- a setup that has completely replaced every CRM he's ever tried. All 410 podcast episodes worth of contacts, hundreds of unanswered inquiries, a live to-do list, follow-up notes from conferences across Europe -- all stored as plain CSVs, updated at the end of every session, accessible from any window or device. No database to maintain. No structure to define upfront. He just talks to it, and it handles the rest. The key detail: because the data lives in GitHub and not inside any one tool, he can swap the underlying model tomorrow and lose nothing. Will counters with his Monday sales workflow inside Claude Cowork -- automatically pulling Zoom transcripts the moment calls end, triaging email threads in Superhuman, resurfacing cold leads, and building context across every deal without a single manual log entry. His whole approach now is getting people onto Zoom calls specifically because the transcript integration is that seamless. From there the conversation opens up into the bigger picture. They get into the hardware pendulum that has swung back and forth in computing for decades -- from Microsoft's software-only bet, to Apple proving integrated hardware wins, to everything moving to the cloud, and now potentially swinging back again toward powerful local machines running open source models that no pricing team, no government, and no platform shutdown can touch. Sam makes the case that the billions currently flowing into data center construction might be the mainframe build-out of this era -- impressive and expensive, and possibly obsolete before it's finished. They also get into the geopolitics: the US government blocking access to a new OpenAI model, what that means for non-US users, and why it only strengthens the argument for open source. Plus the uncomfortable question neither side of the AI industry wants to talk about -- every major model right now is priced below cost and subsidized by venture capital, sovereign funds, and stock market investors. What happens to your workflow, your costs, and your dependencies when that subsidy stops? Honest, practical, and genuinely useful whether you're trying to build a personal AI setup that actually holds up, or just trying to understand where all of this is heading.

22: The hidden truth about AI IPOs and why tech giants are betting big on AI

AI's Booming Year: How SpaceX, Anthropic, and the Rising Tide of Innovation Are Reshaping the FutureA seismic shift is happening in the AI world, and if you're not paying attention, you risk falling behind. SpaceX's imminent IPO, valued at over $1.7 trillion, is just the tip of the iceberg. Meanwhile, companies like Anthropic and OpenAI are redefining valuation norms, and the entire tech landscape is racing to capitalize on AI's explosive growth. If you're curious about how these giants are shaping markets, disrupting industries, and unlocking unprecedented opportunities — this episode is your essential guide.Dive deep into the conversation that unpacks the latest developments, from SpaceX’s secretive plans merging with X AI to the surge in secondary deals and billion-dollar valuations. You’ll discover the strategic shift in AI companies focusing on enterprise over consumer, why the real value lies in data centers and cloud services, and how AI investments are subsidized by traditional giants like Google, Amazon, and even city councils. This episode also explores the broader implications, how innovation, regulation, and usage models could define the next decade of tech evolution.We break down the hype versus the reality: Why many overvalue these nascent companies and what it truly takes to build lasting competitive moats in AI. You’ll learn about new billing models transforming SaaS, the quiet rise of token-based pricing, and how AI is changing everything from CRMs to subscription services, making them more efficient, more scalable, and more aligned with real usage.This conversation matters if you're a founder, investor, tech enthusiast, or simply someone fascinated by the convergence of space, AI, and innovation. With insights from experts immersed in the AI boom, you'll gain clarity on what’s next and how to position yourself for long-term success. Perfect for those who want to stay ahead of the curve in a world where AI's potential is only just beginning to be tapped.Are these valuations sustainable? Will AI disrupt traditional industries, or will it be a slow burn like the taxi industry with Uber? And how will the shifting models—like token billing or AI-powered CRMs, impact your investments or business? Tune in to uncover the answers behind the headlines, with real-world insights and a glimpse of what’s coming next in this exhilarating tech revolution.Why this works: This episode hooks listeners with high-stakes finance and innovation details, sparking curiosity about the future of AI and space tech. It expertly combines concrete examples, like SpaceX’s valuation and secondary deals, with big-picture implications, appealing directly to entrepreneurs, investors, and tech aficionados eager to understand the next wave of disruption. The clear call to explore both hype and reality motivates binge-listening and action-taking.

21: Polsia, Character AI & The Truth About AI-Run Companies

I sit down with Will to dig into the wild, weird, and sometimes worrying state of AI-run businesses right now. We talk about Polsia, the platform claiming to run entire companies for $49 a month, and ask the hard question: are any of these AI businesses actually working? We also get into Character AI, its $2.7 billion Google backing, the lawsuits, the fake doctors, and the pay-per-minute Jesus character. Then we unpack how OpenAI accidentally filled its outputs with gremlins and goblins, why AI is feeding on its own content, and what that means for human language. We round it out with a look at AI trading bots, bug bounty agents, and whether any of this is actually making real money, or just generating hype. Be a guest on the podcast or contact us - https://www.disruptia.news [https://www.disruptia.news] --- CONNECT --- Web3 with Sam Kamani Website: https://www.web3pod.xyz/ [https://www.web3pod.xyz/] Polsia: https://www.Polsia.ai Character AI: https://character.ai Polymarket: https://polymarket.com Kalshi: https://kalshi.com Sora (OpenAI): https://sora.com --- KEY POINTS WITH TIMESTAMPS --- • [00:01] Introduction and catching up, back in New Zealand, diving into the latest in AI • [01:00] Polsia breakdown, $49/month AI-run companies, $9M ARR claims, and what's actually happening on TrustPilot • [03:09] The big debate, should AI be making business decisions, or do humans need to stay in the loop? • [06:12] Algo trading as the exception, why high-frequency decisions work for AI but most business decisions don't • [07:27] AI spam is flooding inboxes, LinkedIn, X, and Telegram, and it's only going to get worse • [09:23] Character AI deep dive, 10 million characters, $60M/year revenue, Google's $2.7B investment, and the fake doctor lawsuits • [13:34] OpenAI's gremlin and goblin problem, how a tweak to avoid real names caused a feedback loop of fantasy characters • [15:08] AI feeding on AI, how generated content is reshaping the internet and possibly human language itself • [16:44] Sora, Seed Dance, and the rise of AI-generated short series in China • [20:31] AI trading bots on Polymarket and Kalshi, separating hype from what actually works • [21:50] Bug bounty agents, a real but modest way to make money with AI at home • [23:00] Open call, if you have ideas for AI agents that make money, reach out and let's build together Disclaimer Nothing mentioned in this podcast is investment advice and please do your own research. It would mean a lot if you can leave a review of this podcast on Apple Podcasts or Spotify and share this podcast with a friend.

20: AI Is Eating SaaS: Why I Just Cancelled Figma, Photoshop & More with podcast hosts Sam Kamani and Will Schmidt

This week I sit down with Will to break down the fastest-moving week in AI yet. I share why I've cancelled most of my team's Figma subscriptions now that Claude Design is here, and we compare it to how Figma disrupted Adobe back in the day. We dig into ChatGPT's image model improvements, the rumoured SpaceX acquisition of Cursor, and why Claude still leads on honesty compared to other LLMs. We also get into something I think about a lot — AI anxiety and whether these tools replace jobs or create new ones. Spoiler: in my own business, I've hired more people after AI than before. Disclaimer:- Nothing mentioned in this podcast is investment advice and please do your own research. It would mean a lot if you can leave a review of this podcast on Apple Podcasts or Spotify and share this podcast with a friend. --- CONNECT --- disruptia.news [https://disruptia.news/] Web3 with Sam Kamani: https://www.web3pod.xyz/ https://www.linkedin.com/in/samkamani/ --- KEY POINTS WITH TIMESTAMPS --- • [00:00] We kick off talking about how fast AI moves — Claude Design dropped just days after our last episode • [01:02] I explain why I cancelled most of my team's Figma subscriptions and why I think Claude Design is the tipping point • [02:47] Will pushes back — Figma still wins for complex, high-fidelity work with big design systems • [03:11] We compare this moment to when Figma disrupted Adobe and Photoshop — history is repeating • [05:31] Canva now has a native Claude integration so you can export straight from Claude into Canva • [07:30] ChatGPT's image model 2.01 finally nails text rendering — a long-standing frustration for everyone • [08:20] I share my Gemini Imagen tip for image generation and how I used to chain it with GPT to remove the watermark • [09:07] Will breaks the news on a potential SpaceX acquisition of Cursor at an $80 billion valuation • [09:46] Claude Code vs Codex vs Cursor — what we're actually using day to day at the agency • [10:44] The biggest remaining problem with AI coding assistants: overconfidence and weird workarounds • [11:17] Claude scores the lowest on the "hallucination confidence" index compared to all other LLMs • [15:48] Will cancelled a personal finance app because he built his own with Claude — and it's better • [17:02] My friend in India uses Claude to replace Xero for personal bookkeeping — it categorises and graphs everything • [18:14] AI anxiety — should you be scared about your job? Our honest take • [19:08] My view: AI doesn't shrink what we create, it multiplies it — I've hired more people since AI than before • [20:27] Why agency and software development is actually a great place to be right now • [22:22] I'm heading to Vegas and Miami for conferences — catching up with AI companies in person

19: How I Use Claude Code and Obsidian to Replace My CRM, My Intern, and My Memory - Will Schmidt and Sam Kamani

In this episode, I sit down with Will to dig into the practical, everyday ways we are using AI to get more done with less effort. We talk about how Will is using Claude Code with Obsidian to build a personal, local CRM that organises itself — no boring admin, no forgotten follow-ups. I share how my 12-year-old built a Google Chrome extension in an afternoon that automates filling out Luma event registration forms, a task I used to pay him to do manually. We also get into the bigger picture: why Meta is on track to overtake Google in ad revenue for the first time, why half the planned US data centres may never get built, and why Claude is quietly winning the enterprise AI race while everyone watches OpenAI. If you want real, no-hype examples of how AI is changing the way we work and build, this episode is for you.  Disclaimer Nothing mentioned in this podcast is investment advice and please do your own research. It would mean a lot if you can leave a review of this podcast on Apple Podcasts or Spotify and share this podcast with a friend.  --- CONNECT --- Disruptia.news  Obsidian: https://obsidian.md/ [https://obsidian.md/]  Orchid.xyz [http://orchid.xyz]  Be a guest on the podcast or contact us - https://www.web3pod.xyz/ --- KEY POINTS WITH TIMESTAMPS --- • [00:00] Sam and Will open with the Allbirds AI pivot and the broader trend of companies rebranding as AI plays • [01:15] Why half of the planned US data centres may never be built due to legal battles, neighbour complaints, and rising electricity costs • [02:00] Will shares how he uses Claude Code for SEO keyword research and competitor analysis via a Node.js app • [02:41] The problem with CRMs — you start using them, get busy, and suddenly the data is three months old • [03:03] What Obsidian is, how it works with markdown files, and why its small team makes serious revenue • [04:48] Using Obsidian as a local personal database — letting Claude Code handle all the organisation and linking • [06:18] Live demo: setting up a brand new Obsidian vault and letting Claude Code populate and connect it • [09:29] Sam's challenge — managing contacts across WhatsApp, Telegram, multiple emails, LinkedIn, and Twitter • [10:33] Will's advice: start simple, paste in Zoom transcripts, and let Claude Code write the notes and to-dos • [13:43] Sam's idea to upload transcripts from 500 podcast episodes and let AI find connections between guests and ideas • [14:04] How Sam's 12-year-old built a Chrome extension to auto-fill Luma event registration forms using Claude • [16:38] Why this kind of tool would have been a SaaS product a few years ago — and why it no longer needs to be • [17:37] The real challenge today is not building, it is distribution, branding, and getting the name out • [19:34] Live Perplexity search: Meta is on track to surpass Google in digital ad revenue for the first time • [20:20] Why the old argument that search intent beats demographic data has flipped — Meta now knows what you want • [21:27] Claude at 30 million users has higher revenue than OpenAI at 900 million — because enterprise and developers choose Claude • [22:37] The AI model arms race: Gemini, Claude, OpenAI, and why competition is great for users • [22:57] Why Google's image AI watermark is ending up in images generated by every other AI model