Should You Trust One AI Model—or Build a Team of Them?
Ryan Hanley's answer to the title question is that one model is not enough yet. He runs several and treats each like an employee with its own job: Claude when a whole team needs the same projects and memory, Kimi K3 from Moonshot for everyday work at a fraction of the cost, GrokBot for agent work inside live systems, and OpenClaw when he wants no guardrails at all. Bradley Flowers brings the other half of the argument from his own hiring process, where the model started hedging on DISC assessments and taking up for the candidate instead of giving him the read he asked for. From there it turns practical. Ryan walks through an insurance prospecting build step by step: anonymize the book, have the AI define five ideal client profiles, build the contact list, then run a seven touch cold drip that adjusts on real feedback. He also explains how an agent found a hundred publishers, ranked them by likelihood, and landed his book deal by emailing five at a time. Bradley has an agent living in his service manager's inbox to win back 60 percent of her day. Scott Howell asks for the 2027 forecast and gets a market that hardens a little, plus a warning about the downward pressure building on agencies that stay out of the tech game. Show Overview 00:00 Intro and sponsors 01:06 Welcome, and the incomparable Bradley Flowers 01:45 New studio, and the property manager on the other side of the wall 02:14 Guest introduction: Ryan Hanley, Finding Peak 03:49 The Mark Walters story: billions shifting under federal investigation 05:11 Smart people do stupid things, and rule one is do not commingle funds 06:14 Regulated versus unregulated, and what you are allowed to say 08:20 Facebook ads for realtors, and the weight loss ad that lasted eight minutes 09:54 Baseball cards, a twelve year old, and a GrokBot arbitrage business 12:28 An Airbnb run on eight programs for a thousand dollars a year 13:51 ChatGPT, Claude, Grok, OpenClaw and GrokBot, sorted by guardrails 14:27 Why Ryan moved to OpenClaw and off Claude 15:56 Running DISC assessments through AI, and the model taking up for the candidate 16:51 Kimi K3, open source, and what a Chinese model actually means 18:21 Treat every model like an employee with its own strengths 19:01 An AI agent living in the service manager's inbox 20:30 Nobody taught AI 101 21:16 Interview the AI about you before you put it to work 21:48 Easy Mode: the book deal, and where the title came from 24:35 How an agent landed the deal, five publisher emails at a time 26:30 The insurance prospecting version of all this 27:31 Five ideal client profiles, HVAC contractors, and a seven touch drip 29:05 You used to think inside the tool, now you just name the goal 30:54 How the book actually gets written: talk to text, then edit 32:34 End every prompt with: ask me your questions one at a time 33:52 The 2027 forecast: harder, but not a hard market 34:21 Do not blame the soft market, and do not take credit for the hard one 34:42 Fifty percent commercial property in a soft market 35:36 PIF and retention are the metrics, not revenue alone 36:31 The downward pressure building on agencies that skip tech 39:07 HawkSoft, Momentum, AgencyZoom, and who is actually moving on AI 41:03 Stop accepting standard development cycles 43:22 Build the feature your CRM does not have 44:55 Why private equity owned platforms do not fund AI 46:03 Hand an agent ownership of AgencyZoom 47:32 ITC, IndieTech, and the conference shakeout 48:08 Rewards come from action, and the sign off