Open Models Handle 90 Percent of Agentic Workflow Steps
Enterprises can optimize costs by routing routine tasks to open-weight AI models.
1Routine Tasks Dominate Workflows
Operational agent workflows consist of hundreds of routine execution steps, including data classification, tool calling, and validation. Analysts Mark Petty and Raymond Paquet suggest that nine out of ten these steps are repetitive. Open-weight models currently handle these routine processes at performance parity compared to closed, proprietary frontier models available today.
2Significant Cost Reduction Potential
Open-weight models offer substantial financial advantages for enterprise users. When compared to the cheapest closed models, open alternatives cost roughly a quarter of the price. Against high-end frontier models used for every step, costs drop to as little as a twentieth. These savings represent a shift in operational expenditure strategies.
3Real-World Efficiency Adjustments
While theoretical savings reach twenty times, practical application shows more modest results. Current open models often exhibit higher verbosity and require more frequent retries during task execution. Consequently, the actual realized saving per task against frontier models is closer to five or six times, which remains a significant operational benefit.
4Data Sovereignty and Control
Beyond cost, open-weight models provide critical data security advantages for enterprises. Sending routine steps to frontier model APIs moves proprietary workflow history outside the corporate perimeter. By utilizing open-weight models, companies retain control over their internal data. This preservation of control is viewed as a primary strategic benefit over cost.
5The Rise of Domain-Specific Models
Enterprises are expected to favor domain-specific open-weight models for their internal operations. These models excel in scenarios where a limited number of highly stable steps are executed millions of times each month. This specialization allows companies to maintain high performance while avoiding the unnecessary expense of using frontier models for simple tasks.
6Value Shift to AI Harnesses
As AI models become commodities, the true value in AI products shifts toward harnesses. These tools manage agent fleets, impose guardrails, and route work to the most appropriate model for each specific job. Effective harnesses ensure that users experience no impact while the system optimizes model selection for cost and performance.
7Frontier Models as R&D Engines
Frontier models will maintain a distinct role within the enterprise AI ecosystem. They function effectively as capital expenditure, serving as the primary engine for research and development. Once these frontier models complete the heavy lifting, open-weight models take over as the operational expense for executing the resulting routine workflow steps.
8Strategic Outlook for 2030
Analysts project that by 2030, 90 percent of enterprise operational agent workflow steps will execute on open-weight models. This transition reflects a broader trend toward commoditizing routine AI tasks. Organizations that successfully implement robust harnesses to manage these model fleets will likely secure a competitive advantage in operational efficiency and security.