AI Strategy: Outcome Pricing and Sovereign Intelligence
Factory CTO Eno Reyes argues that AI valuation must shift from token inputs to outcome outputs. He predicts 80-90% of neo-labs will fail within 18 months and asserts that open models will handle 99% of workflows in three years, necessitating a focus on sovereign intelligence and harness-layer control.
The Shift from Input to Outcome Pricing
The AI industry is undergoing a fundamental valuation shift. Eno Reyes, CTO of Factory, argues that focusing on token costs is a strategic error. The true metric is the price of the outcome. A sophisticated model that resolves a complex task efficiently is cheaper than a cheap model that requires massive token consumption to achieve the same result. This reframing challenges current SaaS pricing models and suggests that AI vendors must prove value through verified results rather than volume.
The Open Source Dominance Thesis
Reyes predicts that within three years, 99% of enterprise workflows will run on open-source models. Frontier models from labs like OpenAI and Anthropic will become niche tools for specific, high-stakes applications such as bio-research or national security. This commoditization of the model layer means that the competitive moat is no longer the model itself, but the infrastructure that deploys it. The "harness"—the software layer that manages state, context, and workflow—becomes the primary source of value and differentiation.
Sovereign Intelligence and Enterprise Risk
A critical risk for enterprises is the loss of "sovereign intelligence." If companies outsource their core workflows to third-party AI providers without retaining ownership of the data and learnings, they risk creating competitors. Reyes emphasizes that businesses must maintain control over their proprietary knowledge. This has led to a resurgence in demand for on-premise solutions, not for security reasons, but for strategic autonomy. Enterprises are wary of model labs that could eventually compete with their clients.
Investment Implications and Market Consolidation
The AI startup landscape is poised for massive consolidation. Reyes estimates that 80-90% of "neo-labs" will fail within 18 months. Investors should focus on companies with durable workflows, proprietary data, and clear paths to margin expansion. The current high valuations of frontier labs are risky because they assume sustained high margins in a market that is rapidly commoditizing. The future belongs to companies that can efficiently allocate intelligence, whether through open models or specialized internal tools, while maintaining control over their core intellectual property.
Key insights
-
The cost of AI should be measured by the price of the outcome, not the cost of the tokens used. A higher-quality model that solves a problem efficiently is often cheaper than a lower-quality model that requires extensive iteration.
Impact: This shift forces AI vendors to focus on efficiency and value delivery, potentially disrupting current subscription-based pricing models and favoring outcome-based contracts.
-
Open-source models will handle 99% of enterprise workflows within three years, while frontier models will remain niche for specialized, high-stakes tasks. This commoditization reduces the strategic importance of proprietary frontier models for general business use.
Impact: Enterprises can significantly reduce AI costs by adopting open models, shifting competitive advantage to the integration and orchestration layer rather than the model provider.
-
Sovereign intelligence is a critical strategic concern. Enterprises must retain ownership of their AI learnings and data to avoid creating future competitors or losing leverage in negotiations with AI providers.
Impact: This drives demand for on-premise and hybrid AI solutions, allowing companies to maintain control over their proprietary knowledge and workflows.
-
The "harness" layer, which manages state, context, and workflow integration, is the new competitive moat. The model itself is becoming a commodity, while the ability to effectively orchestrate models within a business context is the key differentiator.
Impact: Investors and founders should focus on building robust harnesses and integration tools rather than developing new models, as this is where the long-term value lies.
-
80-90% of AI startups will fail within the next 18 months. Only those with durable, proprietary workflows and clear paths to margin expansion will survive the consolidation phase.
Impact: This high failure rate suggests a need for selective investing, focusing on companies with strong unit economics and defensible business models rather than just technological novelty.
Action items
-
Re-evaluate AI pricing models to focus on verified business outcomes rather than token consumption. Implement metrics that track the value delivered by AI solutions to justify costs.
Impact: This aligns AI spending with business value, improving ROI and justifying higher budgets for high-impact AI applications.
-
Develop a strategy for sovereign intelligence by retaining ownership of AI learnings and data. Consider on-premise or hybrid deployments to maintain control over proprietary knowledge.
Impact: This reduces dependency on third-party AI providers and protects the company's competitive advantage by ensuring that AI improvements benefit the business directly.
-
Invest in the harness layer, focusing on stateful context management and workflow integration. Build or acquire tools that effectively orchestrate models within the business context.
Impact: This creates a defensible moat by improving the efficiency and reliability of AI deployments, regardless of the underlying model used.
-
Assess the durability of AI startup investments by evaluating their workflow specificity and margin expansion potential. Avoid companies that rely solely on technological novelty without a clear path to profitability.
Impact: This reduces investment risk by focusing on companies with sustainable business models and clear value propositions, rather than those likely to fail in the consolidation phase.
-
Implement robust verification frameworks for AI outputs before scaling deployment. Define clear criteria for what constitutes a successful outcome in each domain.
Impact: This ensures the reliability and trustworthiness of AI solutions, reducing the risk of errors and building confidence among stakeholders.
Quotes
“I see a world where the smartest model is actually the cheapest.”
“In three years, 99% percent of workflows are going to be done on open models.”
“I think it could be 80 to 90% of Neolabs die in the next 18 months.”