Running a Company Through AI Agents: Ian Karnell @ VastAdvisor
AI has made building software nearly free. For most industries, that's the whole story — ship faster, iterate faster, done. In wealth management, it isn't. The moment AI touches anything client-facing, the question stops being "what can this do" and becomes "can you show a regulator exactly what it did, and why." Ian Karnell has spent the last few years living on both sides of that line. He's co-founder and CEO of VastAdvisor, which he built after co-founding and selling Truelytics to Envestnet. But this conversation isn't really about VastAdvisor's product — by the hosts' own admission, they barely get to it. It's about what Ian has built inside his own company: a roughly 40-agent internal AI system, segmented into its own board of directors, executive functions, and departmental "guilds," handling everything from live code fixes to compliance review. Along the way, Ian makes the case for why governance and telemetry — not features — are becoming the real differentiator for anyone deploying AI in a regulated industry, and why he thinks loyalty to any single AI model won't mean much in a couple of years. What You'll Learn How VastAdvisor's holding-company structure (VastAssembly) separates core IP ownership from the operating businesses built on top of it Why Ian moved off month-to-month contracts once he had proof of product-market fit, and how his earliest customers got locked into legacy pricing for staying early The internal structure of VastMind, VastAdvisor's ~40-agent system, including its "board of directors" and function-specific guilds How the team catches and fixes production code errors in real time, with human review calibrated to the severity of the fix Why Ian considers AI telemetry and observability — not marketing-content compliance — the harder unsolved problem for regulated AI deployments How VastAdvisor manages token costs across enterprise and individual AI subscription plans Ian's take on whether loyalty to any one AI lab or model will still matter in two to three years Key Takeaways Governance is becoming the product. → In a regulated industry, proving what your AI did — and why — is starting to matter as much as what the AI can do. Product-market fit is a spending decision, not a talking decision. → Ian's advice to founders is blunt: paying customers, not user interviews, are the real signal that something works. The agent swarm isn't a demo — it's the org chart. → VastAdvisor runs internal functions like compliance review and code fixes through a segmented system of dozens of AI agents, not a single assistant. Token spend is a strategic line item, not overhead to minimize. → Ian treats AI costs the way most companies treat payroll: something to lean into, not shrink. Nobody is fully loyal to one model anymore. → VastAdvisor is built to be LLM-agnostic on purpose, because the labs themselves are moving too fast to bet the company on just one. Why This Episode Matters Most AI conversations in wealth management are about customer-facing tools — chatbots, portfolio commentary, marketing copy. This episode is a reminder that the bigger shift may be happening behind the scenes, inside the companies building those tools. When a 25-year veteran of the space says his own company runs largely through an agent swarm, and that the harder problem isn't building the AI but proving it behaved correctly, that's a signal about where scrutiny — from regulators and from clients — is heading next. For advisors and wealthtech leaders evaluating AI vendors, that reframes the diligence question. It's less "what can this do" and more "can you show me that it did that."