The Intersection Magazine Podcast

Production Title: Episode 2: The Enterprise Tech Stack: LLMs, RAG, MCP, and Sub-Agents

Subtitle: A Strategic Implementation Roadmap for Enterprise Technology LeadersEpisode Summary: Designed specifically for technology leaders, this episode provides a strategic roadmap for building and deploying an autonomous AI workforce. We break down the core layers of the modern AI stack—Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), and Model Context Protocol (MCP)—while exploring concrete agentic workflows, sub-agent security, and practical implementation strategies.

Production Title: Episode 1: From Per-Seat Labor to Outcome-Based AI Operations

Subtitle: How MCP Protocols and Sub-Agent Swarms Are Redefining Enterprise BPO Models

Episode SummaryIn this episode, we examine how autonomous AI agent swarms and Model Context Protocol (MCP) are transforming traditional enterprise labor. We analyze the economic shift from human labor arbitrage to tech-enabled automated outcomes, evaluate the transition from per-seat/FTE billing to outcome-based AI pricing, and discuss how Business Process Outsourcing (BPO) providers are evolving into AI-managed service providers

In this deeper cut of the episode of The Architectures of Intelligence, Alex and Devon go beyond the surface of the Defensibility Mirage to examine where real competitive advantage actually lives in the age of Service-as-Software.

As models and infrastructure become pure commodities, the episode breaks down the durable moats that matter:

• Workflow ownership — controlling the full operational sequence so switching costs become structural
• Proprietary context and outcome-labeled data — the messy, compounding record of what actually worked
• Trust architecture — structural guardrails, semantic validation, and auditability that give systems the right to act
• Deep integrations that embed the product into the customer’s operational reality

The conversation also maps the 2026 AI agent framework landscape — LangGraph for production-grade stateful workflows, CrewAI for rapid multi-agent crews, the Microsoft Agent Framework, and the OpenAI Agents SDK — while making a critical strategic point: the frameworks themselves are rapidly becoming infrastructure. The lasting advantage lies in the domain logic, evaluation systems, and proprietary orchestration built on top of them.

Clear-eyed and practical, this expanded episode is essential listening for founders and operators who want to build systems of action rather than just smarter tools.

Episode Description

In this deep-dive episode of Agentic Intelligence, we unpack one of the most practical frameworks in enterprise AI today — the five levels of AI agent maturity.

We move beyond high-level slides to examine the technical mechanisms behind each level, the real operational consequences of staying too long at lower levels, and how the maturity curve actually plays out in high-stakes industries.

You’ll hear clear definitions and working examples across:

• Finance — fraud detection, credit decisions, collections, and loan onboarding
• Healthcare — prior authorization, care coordination, and population health
• Retail — returns, inventory allocation, dynamic pricing, and personalization

We cover:

• The seven foundational traits that separate true agents from glorified chatbots
• Why Level 1 and Level 2 create hidden operational drag even when they look successful
• How Level 3 agents begin to own multi-system processes
• Why true Level 4 utility-based agents require explicit business-owned trade-off models
• What Level 5 learning agents actually need to improve safely (and why most organizations aren’t ready)
• The strategic mistake of chasing one “super agent” versus building a governed portfolio of specialized agents

Whether you’re evaluating current automation, planning your next wave of agent deployments, or trying to separate real progress from vendor hype, this episode gives you a sharper diagnostic framework and clearer language for the conversations that matter.

Ideal for technology leaders, transformation executives, and practitioners working in regulated or complex operational environments.

Key Takeaways

• Most organizations are still operating primarily at Levels 1–2, with selective Level 3 pilots.
• Level 4 requires explicit, living utility functions owned by the business — not just clever prompts.
• Level 5 only works with clean feedback loops, strong governance, and controlled exploration.
• The winning architecture is a distributed portfolio of specialized agents, not one general-purpose super agent.
• Industry patterns differ: Finance pushes hardest into Levels 4–5; Healthcare moves more cautiously; Retail has rich data but struggles with consistency.

In the final episode of The Architectures of Intelligence, Alex and Devon dismantle one of the most persistent illusions in AI: the belief that the smartest model or the most powerful infrastructure creates lasting competitive advantage.

As GPU computing becomes a pure commodity and upstream CapEx balloons into a $600 billion annual question, the real battle shifts to the application layer. The winners won’t be the teams that train slightly better models. They’ll be the ones building Service-as-Software — systems that don’t just assist humans, but execute the work and deliver finished business outcomes.

Alex and Devon explore:

• Why seat-based SaaS pricing collapses when software does the labor
• Concrete examples of Service-as-Software in action (Sierra, Decagon, Harvey, Tennr, Flock Safety, and more)
• The true sources of defensibility: workflow ownership, proprietary context and outcome-labeled data, deep integrations, and trust architecture
• How today’s AI agent frameworks (LangGraph, CrewAI, Microsoft Agent Framework, and others) are themselves becoming commodity infrastructure
• Why the durable moat lies in owning the customer’s operational workflow rather than the underlying intelligence

Clear-eyed, practical, and free of hype, this episode is essential listening for founders, product leaders, and operators building in the age of agentic AI.

Hosts: Alex & Devon
Series: The Architectures of Intelligence

Build-vs-buy heuristics and the rise of the hybrid application developer.

A common pitfall for AI startups is "over-engineering"—spending millions of dollars and precious developer hours building custom infrastructure that does not actually differentiate their product. This episode tackles the build-versus-buy dilemma for AI product builders using a structured decision framework. We explore the critical role of "AI Application Infrastructure" (the middleware layer: agent orchestration, RAG, evaluation, and guardrails)and argue why a modular, hybrid architecture is the most defensible default for growth-stage startups. By buying commodity infrastructure, product teams can dramatically reduce time-to-value and focus their design talent entirely on user workflows.

SaaS 2.0, "Service-as-Software", and the collapse of traditional user-licensing.

The SaaS 1.0 era was built on "systems of record" operated by humans, monetized through seat-based subscriptions. But as AI transitions applications from tools into "systems of prediction and execution," we enter the era of SaaS 2.0 (Service-as-Software)—where companies sell completed labor rather than software licenses. In this episode, we discuss how selling completed workflows dramatically expands the software market from a $1 trillion software market into a $4.6 trillion+ labor market. We also examine the massive design and economic shift this causes: when software does the job of an entire team, the "per-seat" business model collapses, forcing product designers to build transaction-based interfaces aligned with the value of completed work.

Moving beyond raw capabilities to build invisible, outcome-oriented software.

For decades, technology design was about building "shovels"—giving users tools to perform tasks manually. In this episode, we explore the foundational split in the modern AI stack: Infrastructure enables intelligence, while Applications package intelligence into outcomes. We unpack the structural differences between infrastructure companies (whose inputs are developer problems and outputs are raw capabilities) and application builders (whose inputs are user problems and outputs are completed outcomes). The core design challenge of the AI era is shifting away from exposing raw, intimidating model parameters (like open-ended chat boxes) toward crafting "invisible AI" that lives silently inside highly specialized, end-to-end user workflows.

In this month’s issue, we have several ideas we are working on under the broad theme of investment.We start of with a piece from one of our favorite companies-NFX.The piece talks about the inevitability of superintelligence and what role this might play in the world of venture capital.The writer argues that the most important part is the creation of the ‘data highway’ that will underpin the union between superintelligence and the unavoidable changes that are coming. What we have seen is the focus on coding as the first wave, followed by the entrance into legal. The next challenge relates to biology and the ability of superintelligence to solve complex regulatory issues such as clinical trials. The suggestion of the article is then to define which problems are most worth solving.Our second section covers an area we have dealt with before although at the time we tackled it on its own. This time we look at Venture Capital and compare it to Corporate Venture Capital. We also loop in Angel Investing. We are interested in discovering what options exist for founders who are looking for capital.How do the interests of each group come into play? What kinds of timelines are in play? How much of the company might the founder have to give up or to offer?Next we have a Q and A with Daaki Joshua. Mr. Daaki has seen both sides of the table. A founder who has actually built something but also served on the investor side. Josh Daaki also sheds some light into his industry covering web design and graphic design as well as his work with StartHub Africa.In our Social Watch Section, we share an interesting conversation about how AI has invaded the wall street by open sourcing entire workflows.Our book recommendation is hashtag#Hooked by Nir Eyal. The book is written to help us understand what makes some products more habit forming than others. It is designed to help you build better but with a useful section within about the ethics of habits and addictions in product design too.

In the month of July we complete the section we started last month about the union between SaaS and #ArtificialIntelligence. As we mentioned before, there is nothing in the history of computing that has been as profitable as Software as a Service. We feature quite a lot of material from the NFX team and we think that this attempt to look into a future post #SaaS is worth reading. We are experiencing a major shift in how we work with major changes taking place use of AI in areas such as robotics. You could say we are at the same point in history where traditional implements were replaced by Agricultural Machinery and Industries rose to transform and create manufacturing.

We also share a piece from ideas that were developed by the NFX team-Network Effects. We share the 16 #networkeffects (phenomenon where the value of a product, service, or platform increases as more people use it). This piece describes the various manifestations of these effects. How do these network effects work? How do they relate to one another? How do we make best use of these effects?

In our Q and A, we have a conversation with Roland F Ganafa an important player in the AI space in Uganda. We are interested in his work and projects that he has undertaken. We are interested in finding out how he has navigated the business space as well as his interactions with government. Given the rapid changes that are taking place in the field of Artificial Intelligence we also have a conversation about how to position the country to be in a more competitive position.