Summary
Guest episode with Matt Fitzpatrick (CEO, Invisible Technologies; former McKinsey global head of Quantum Black Labs) on why every company must become AI-native in 2026. Core thesis: not all industries are equally impacted — media, legal, and BPO face structural disruption while oil & gas and real estate see less change. The key debate is whether AI-native startups get distribution before incumbents build capability. The episode dives deep into custom benchmarks vs. public benchmarks, the Klarna contact center rollback, and concrete AI use cases including Charlotte Hornets draft prep (computer vision on player movement), Lifespan MD (concierge medicine data platform), and enterprise workflow automation.
Key Segments
- [00:01] Intro framing: every company must become AI company in 2026 or face disruption
- [00:05] Matt’s framework: not all sectors equally impacted; media, legal, BPO most disrupted; oil/gas less so
- [00:07] Key tension: do startups get distribution before big companies build tech?
- [00:10] Legal/accounting as case studies for AI disruption; Harvey as AI-native legal challenger
- [00:13] Klarna contact center post-mortem: moved to fully agentic, then rolled back to humans; lesson is hybrid approach required
- [00:17] CEO playbook: pick 2-3 high-impact use cases, RFP to vendor compensated on outcomes, don’t let 1000 flowers bloom
- [00:21] AWG on custom benchmarks: need thousands of narrow industry-specific evals, not just broad public benchmarks
- [00:25] Bloomberg GPT cautionary tale: proprietary models leapfrogged by generalist frontier models within months
- [00:29] Charlotte Hornets case study: custom CV model for draft prep analyzing player spatial movement patterns
- [00:32] Lifespan MD: HIPAA-compliant multi-tenant data platform for concierge medicine practices
Notable Claims
- Models improved 50-100% on most benchmark dimensions in last 3 years
- Most enterprise AI adoption still stuck at pilot stage; few at scale deployment
- Klarna claimed AI handled 2.3M calls/month (700 FTE equivalent) in month one, then fully rolled back
- Custom enterprise evals are the bottleneck, not model capability
Guests
- Matt Fitzpatrick — CEO of Invisible Technologies, former global head of Quantum Black Labs (McKinsey)
- Peter Diamandis — Host, XPRIZE founder
- Salim Ismail — Co-host, Exponential Organizations author
- Dave (DB2) — Co-host, entrepreneur
- Alex Wezner-Gross (AWG) — Co-host, physicist/technologist
RDCO Mapping
- Custom evals: RDCO should build domain-specific benchmarks for its operational tasks rather than relying on general model evals
- Enterprise AI adoption: confirms that AI-native startups (like RDCO’s model) have structural advantage over legacy companies trying to bolt on AI
- Hybrid human-AI: Klarna rollback validates keeping humans in the loop for edge cases; relevant to RDCO agent design
- Outcome-based pricing: Matt’s RFP-for-outcomes model worth considering for RDCO service positioning
Related
- ai-landscape
- ai-benchmarks
- exponential-organizations
- enterprise-ai-adoption