How AI Stacks are rewriting the Rules of Business

The podcast discusses the evolving landscape of AI, focusing on the shift from isolated AI models to integrated AI software stacks that are transforming enterprise operations. A key theme is the emergence of "systems of intelligence" that combine data, processes, and tacit knowledge, enabling dynamic interactions through LLMs, agents, and co-pilots, in contrast to traditional business intelligence tools. This shift is driving changes in business models, including a move from consumption-based to outcome-based pricing, and challenging both vendors and buyers to rethink how AI is adopted and integrated into existing systems.

Another major focus is the competitive dynamics in AI infrastructure, particularly the dominance of Nvidia in AI hardware and the efforts by AMD and others to provide alternatives through open co-innovation and ecosystem development. The discussion highlights the importance of engineering velocity, open-source collaboration, and collective action among AI companies to avoid over-reliance on a single vendor. Additionally, concerns around data governance, sovereignty, geopolitical influences, and the economic sustainability of the AI market are explored, along with the potential for distributed AI systems and the economic shift from labor to AI-driven token-based costs.

• What if you built a micro "system of intelligence" for a niche business process?

  • Move: Identify a high-friction, repeatable business function (e.g. invoice dispute resolution, customer onboarding QA) and build a lightweight agent stack using an LLM API, structured prompts, and a simple feedback loop to codify tacit knowledge.
  • Why Now?: AI tooling (prompting, agentic frameworks, RAG) is stable enough for narrow deployments, and enterprises are actively seeking outcome-based automation - not just dashboards.
  • Expected Upside: You create a monetizable, reusable module that demonstrates outcome-based value (e.g., "reduces onboarding errors by 40%"), positioning you as a builder of intelligence-layer software rather than just tools.

• What if you priced your AI tool based on outcome, not usage?

  • Move: Refactor your pricing model from API calls or seats to a milestone- or result-based fee (e.g., $500 per validated contract analyzed, $2k per process optimized), using FDE-style delivery with clear KPIs.
  • Why Now?: The market is shifting from consumption to outcome pricing, and early adopters can differentiate by aligning economic value with client ROI - not just compute costs.
  • Expected Upside: Higher perceived value, stronger client retention, and faster sales cycles due to risk-sharing - especially effective for solopreneurs offering precision AI services in verticals like legal, HR, or logistics.

• What if you offloaded AI compute to on-prem or edge token generation for clients?

  • Move: Offer a lightweight, self-hostable token runner (e.g., using Llama 3 8B or Phi-3) for clients concerned about data sovereignty, cost, or latency, bundling it as a managed micro-service.
  • Why Now?: Token costs are falling (90% YoY drop), small models are improving, and demand for distributed AI is rising - especially with clients wary of cloud lock-in or vendor learning their data.
  • Expected Upside: You unlock premium pricing for privacy-aware deployments, reduce client dependency on OpenAI/Nvidia, and position yourself as a builder of sovereign, edge-capable AI stacks - tapping into geopolitical and cost-driven demand.

• Adopt outcome-based pricing models for your software products by aligning fees with measurable client results, moving beyond traditional per-user or per-seat pricing.
• Build or integrate narrow, milestone-driven automation modules (e.g., agent workflows) that codify specific business processes, enabling predictable delivery and cost tracking.
• Invest in on-premises or private AI inference setups (e.g., local token generation) to reduce long-term dependency on external API providers and manage escalating usage costs.
• Focus on creating systems of intelligence by embedding tacit knowledge and internal workflows into your applications, using LLMs or agents to enable dynamic interaction and learning loops.
• Prioritize simplicity and determinism in your AI features - design reproducible, reliable outputs and avoid overloading your product with unproven capabilities during early adoption phases.