Chapter 1 established that AI is a sandpile – capabilities accumulating from thousands of sources, approaching a critical state that no one can predict or control. Disintermediation isn’t just a prediction; it’s baked into the physics of the model.
But why? What makes AI inherently resistant to centralized control? And how do individual grains of sand actually form?
Why Disintermediation Is Inevitable
The traditional SaaS moat depends on two things: data lock-in and integration complexity. You stay with Salesforce because your data lives there and connecting everything else is painful.
AI breaks both mechanisms.
Data lock-in fails because customers own their data. APIs provide equal access to that data for internal agents, platform vendor agents, and third-party agents. A Salesforce agent has no inherent advantage over an AWS agent or an open-source agent accessing the same data.
Integration complexity fails because LLMs can work across systems. They understand context, translate between data models, and orchestrate workflows that previously required custom integration code. The complexity that protected incumbents becomes irrelevant when AI navigates it automatically.
The last moat is governance. “Only our agents can safely access your data” becomes the final defensive position. It’s why Salesforce led their Agentic rollout with the governance layer. But third-party governance platforms are already emerging – companies like Credo AI and BigID building cross-platform oversight. Microsoft announced Agent 365 as a “control plane for AI agents across Microsoft, third-party, and open-source platforms.”
If governance itself gets disintermediated, even the last moat collapses.
But the deeper reason is structural. Capability discovery is happening everywhere simultaneously. Thousands of organizations running POCs, experimenting, solving specific problems. No single vendor can monopolize this process. No vendor can prevent capability replication once patterns are identified. No platform can maintain technological advantage when infrastructure commodifies.
The sandpile grows through “mindless sprinkling” – grains added from every direction, forming networks of instability that transcend any single platform. This is the physics. Fighting it is futile.
Two Levels of Disruption
Understanding why disintermediation is inevitable requires distinguishing between two levels of transformation:
Level 1: Inside Your Company
This is what consultants solve for and what most organizations focus on:
- Workflow redesign and process optimization
- Change management and adoption
- Configuration decisions and system integration
- Training, governance, and capability building
McKinsey found that 55% of high performers redesigned workflows versus only 21% overall (McKinsey 2025). This matters. Configuration is real. These are sophisticated process improvements that deliver genuine competitive advantage.
But Level 1 optimization doesn’t protect you from Level 2 disruption.
Level 2: The Universal Sandpile
This is the industry-level transformation that no single company controls:
- Capabilities emerging from thousands of sources simultaneously
- Infrastructure commoditizing as competition drives down costs
- Business models becoming obsolete overnight
- Value chains being disintermediated
The paradoxes from Chapter 1 exist because companies are trying to optimize Level 1 while fearing Level 2. The consultancy data measures Level 1 dynamics – what works for workflow optimization. The behavioral data – fear, speculative investment, workforce positioning – reveals Level 2 awareness.
You can be perfectly optimized internally (Level 1) and still get obliterated by external disruption (Level 2) because AI commoditized your value proposition.
Consider a digital marketing agency that masters AI-assisted content creation, campaign optimization, and client reporting. They achieve 40% efficiency gains. Level 1 success. But if AI enables clients to do this work in-house at 10% of the cost, the agency’s entire business model disappears. Level 2 disruption.
This is why disintermediation is inevitable. Individual companies can optimize their sandpile configuration, but they can’t prevent the universal sandpile from reaching criticality.
The Capability Lifecycle
If disintermediation is inevitable, how do individual capabilities – grains of sand – actually form? They follow a predictable lifecycle:
Phase A: Process Improvement
Most AI work starts here. An organization runs a POC to improve an existing process. They achieve 20-50% efficiency gains – real but modest. A task that took 10 hours now takes 6. A report that required manual compilation now generates automatically.
This alone may rarely justify significant investment. But something important is happening: the organization is learning where AI works and where it doesn’t. They’re mapping the terrain.
Phase B: Capability Discovery
In the course of POC work, something unexpected emerges. The AI doesn’t just improve the process – it enables something that was impossible before.
This is the critical transition. Process improvement asks: “How do we do this faster?” Capability discovery asks: “What becomes possible that wasn’t before?”
The difference is profound. Process improvement thinking leads to incremental gains. Capability thinking leads to 10-100X transformations – or entirely new business models.
A new capability isn’t just about better automation. It operates at a level of abstraction that wasn’t accessible before. It creates new feedback loops, unlocks downstream value, or changes organizational behavior in ways that compound over time.
Phase C: Optimization and Deployment
Once a new capability is discovered, the business case becomes straightforward. The ROI easily justifies usage costs. Organizations adopt rapidly.
New businesses emerge to deploy and optimize these capabilities. To maximize margin, they focus on reducing infrastructure costs. Initial discovery uses expensive models and prompt-based implementations (fast to iterate, slow to execute). Optimization migrates to cheaper models or more efficient code.
As this happens across thousands of capabilities simultaneously, competition drives down infrastructure costs. The infrastructure vendors who built the sandpile during accumulation phase face margin pressure during deployment phase.
Lower usage costs then unlock the value for the next set of capabilities.
This resolves the first paradox: infrastructure built on speculation, justified later by deployment volume.
Multiple Paths to 10-100X
Capability discovery isn’t a single pattern. There are multiple paths to transformational value:
Path 1: The Meta-Capability
Creating new organizational feedback loops that didn’t exist before. The capability doesn’t just automate a task – it shapes behavior, embeds discipline, and enables higher process maturity. Value compounds over time as the organization changes how it works.
Path 2: Massive Parallelization
Daniel Englebretson describes running 3-5 meta-agents, each coordinating 3-5 sub-agents, executing work in parallel. What took a weekend now takes 7 minutes. This isn’t about creating new capabilities – it’s about orchestrating existing AI tools at scale in ways that were impossible before.
As he puts it: once you convert analog work to digital workflows, you can run them in parallel, iterate improvements, and collapse entire value chains. The 100X comes from doing at scale what was previously limited to serial human effort.
Path 3: Value Chain Collapse
AI enables organizations to collapse multi-step value chains that previously required different specialists, departments, roles, or vendors. Work that used to flow through sequential handoffs can now execute end-to-end with fewer intermediaries.
A legal team that previously needed separate researchers, analysts, and brief writers can now have one attorney orchestrate AI agents across the entire workflow. A client that previously hired marketing agencies for strategy, creative, and execution can now handle the entire chain in-house with AI assistance.
This isn’t just efficiency – it’s elimination of structural costs. The 10-100X comes from removing entire layers of coordination overhead, handoff delays, and intermediary margins.
This is Level 2 disruption: existing AI capabilities making entire categories of intermediaries unnecessary.
The Discovery Imperative
The capability lifecycle reveals why the current moment matters.
During accumulation phase, organizations are discovering which grains of sand will form the network of instability. The ones discovering capabilities now gain first-mover advantage. They understand the patterns. They’ve mapped the terrain.
When the avalanche comes, it won’t reward the organizations that waited. It will reward those who spent the accumulation phase learning where the valuable grains exist.
The question isn’t whether to participate in capability discovery. The question is whether you’re finding process improvements or true capabilities – and whether you recognize the difference.
References
- Daniel Englebretson, “Friction-Focus Framework: Actionable AI Strategy,” LinkedIn (2024) – https://www.linkedin.com/pulse/friction-focus-framework-actionable-ai-strategy-daniel-englebretson-xmpae
- McKinsey, “The State of AI: How Organizations Are Rewiring to Capture Value” (2025)
- Microsoft, “Introducing Microsoft 365 Copilot and Agent 365” (2025)