---
title: "The model is not the moat: Where enterprise value moves in the agent era"
description: "AI models are commoditizing fast. The durable advantage of the next decade belongs to firms that own the workflow, earn trust, and grow human judgment alongside their agents."
date: "2026-10-28"
author: "SuperOrgs"
category: "Insights"
source: "https://www.superorgs.com/blog/the-model-is-not-the-moat"
---

# The model is not the moat: Where enterprise value moves in the agent era

AI models are commoditizing fast. The durable advantage of the next decade belongs to firms that own the workflow, earn trust, and grow human judgment alongside their agents.

In November 2021, GPT-3-level quality cost $60 per million tokens. Three years later, a model with the same benchmark score cost about six cents. That's a roughly 1,000x drop, by [Andreessen Horowitz's measure](https://a16z.com/llmflation-llm-inference-cost/).

When the core input to a technology gets that cheap that fast, the competitive advantage doesn't go to whoever has the most of it. It goes to whoever puts it to work best.

In the agent era, enterprise value moves away from the model and toward three things a company has to build for itself: ownership of the workflow, trust with customers and regulators, and human judgment that grows alongside its agents.

## The numbers are compounding

IDC forecasts that the active agents organizations use will reach [2.5 billion by 2030](https://www.idc.com/resource-center/blog/the-agent-economy-is-scaling-faster-than-it-can-be-metered/), nearly 80 times the 2025 number, completing 459 trillion actions a year.

Agents will multiply per workflow, not per user. A single knowledge worker in 2030 will run many agents at once, and those agents will spawn others.

Yet in [McKinsey's 2025 State of AI survey](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-2025), no more than 10 percent of respondents reported scaling AI agents in any one business function, while 62 percent said their organizations were at least experimenting with them. The [2026 edition](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai) found 40 percent of respondents at companies with more than $1 billion in revenue now scaling agents. The gap between the forecast and what most companies run today is the opportunity, as well as the risk.

## Why the model isn't where value lands

Three signals point the same way:

1. From November 2021 to November 2024, the cost of the cheapest model above a fixed quality bar fell roughly tenfold every year.
2. Open-source agent harnesses such as OpenClaw, Hermes Agent, and Kilo Code are running tens of trillions of tokens a month in production.
3. Large enterprises increasingly put a model router in front of several providers, and the router, not any one model, is becoming the asset they protect.

The pattern shows up in pricing, too. Since May 2025, Salesforce has sold Agentforce on [Flex Credits](https://www.salesforce.com/news/press-releases/2025/05/15/agentforce-flexible-pricing-news/), a consumption model where customers pay for the actions agents perform, such as updating a record or resolving a case. In April 2026 it launched [Headless 360](https://www.salesforce.com/news/stories/salesforce-headless-360-announcement/), which makes everything on the platform an API, MCP tool, or command that agents can use. Per-user licenses are still sold, but value is moving from access to outcomes.

## Where the margin goes

If the model isn't the moat, what is? The durable margin goes to firms whose human and agent capacity has been *jointly designed*, not just jointly accumulated.

It shows up in three places:

- **Workflow ownership.** Own the imperative, the data, the customer relationship, and the feedback loop, and models become interchangeable inputs. Vendors rotate. Workflow stays.
- **Trust capital.** Trust is the currency of agent-mediated commerce. Firms that already hold it with customers, regulators, and partners have a lead that agent-native rivals must rebuild from scratch. Agents can make a gesture faster and more personal. They can't replace it.
- **Compounding human capability.** A person who coaches three agents this quarter can coach thirty next year. Agent capacity is cheap to add. Good human judgment isn't. Firms that hire, train, and retain for that asymmetry will get more from both.

## The firm that wins

The winning firm of the next decade won't have the most agents. It won't have the most people. It will be the firm that designed the seam between the two most rigorously, with named owners, a runtime that can be audited, and metrics that match how work actually gets done now.

We call that firm a **SuperOrg**. The next 36 months will decide who builds one by design, and who backs into one by accident.

## Read the full thesis

The white paper lays out the evidence, the five-layer architecture, the four principles of the SuperOrg, and the open questions we think every leader should be asking.

- [Download the white paper](/future-of-enterprise-work)
- [Read where agents break](/blog/ai-agent-failures)
