Does Your AI Model Choice Even Matter? Why the AI Harness is the More Valuable Investment

At Ridgeline, we’ve made a deliberate bet: in the age of AI, the model will not be the enduring differentiator. Instead, it will come from the system built around that model.

Microsoft CEO Satya Nadella recently described the risk in refreshingly direct terms. He argued that companies should retain the data surrounding every use of a model and keep their harness, context, and memory separate from the model itself. Without that control, Nadella argues that a company has essentially outsourced its thinking. The point is not that companies should stop using frontier models. It is that the model should remain a replaceable source of intelligence rather than becoming the permanent repository of the company’s memory and reasoning.

Peak is Ridgeline’s answer to that risk. The model itself is not the source of our long-term advantage. Ridgeline retains and governs the usage, context, and reusable assets. Because those layers remain separate from the model, Ridgeline can continue to adopt the best available intelligence without surrendering the operating knowledge that makes that intelligence useful. Peak turns repeated use of AI into shared knowledge that is specific to Ridgeline and increasingly difficult to replicate.

Control determines where the value created through AI use accumulates. As model capability becomes broadly accessible, durable advantage shifts from model access towards control of the context, decisions, corrections, and outcomes generated around it. A company that leaves those signals fragmented across individual conversations, local configurations, and vendor-specific systems may receive useful answers without becoming meaningfully smarter as an organization.

‍

Bring AI to Customer Data Safely

Peak allows Ridgeline to bring AI to sensitive customer data without surrendering control over how that data is accessed, interpreted, or acted upon. Employees can use Peak to investigate customer issues, implement new customers, analyze product data, and provide support within a governed environment. 

Peak inherits the permissions and customer-segregation rules of Ridgeline. This ensures that each user and agent sees only the information required for the job and can take only authorized actions. At the same time, Peak records the context retrieved, tools invoked, approvals obtained, outputs produced, and actions taken, creating an auditable trail for oversight. The same audit trail also gives Ridgeline the evidence needed to evaluate performance, identify risks, and continuously improve Peak. 

Peak’s advantage is that it can use the data entrusted to Ridgeline while retaining control over access, governance, and auditability. Vendor harnesses must generalize across many companies and operating models. Peak can instead encode Ridgeline-specific context, permissions, customer boundaries, and controls directly into the execution layer. That is the difference between a generic execution engine and an enterprise operating layer. 

‍

Tailor the Harness for How the Business Works

Peak is centrally managed and cloud-based, meaning improvements can be deployed across the company without requiring every user to change behavior, update local configurations, or learn a new workflow. 

For example, recently our team identified a way to make tool calls approximately 70% more token-efficient. After validating the change against real Ridgeline workflows, we deployed it centrally so every Peak interaction automatically inherited the improvement, with no changes required from users.

While Anthropic or OpenAI can make changes to their harnesses that also lead to platform-level improvements (like tool routing, caching, agent loops, etc.), those improvements are shared (including any Ridgeline competitors who also use the vendor’s harness) and not Ridgeline specific. When we make an improvement to Peak, we do so because of a Ridgeline-specific workflow that is validated against real Ridgeline work. A default harness wouldn’t have the data to make those changes, wouldn’t be able to validate the change against Ridgeline-specific jobs, and would not automatically enforce it across independently configured agents and users. Vendor improvements to a generic harness raise the baseline, but Peak improvements create proprietary leverage. 

‍

Shift AI From Reactive to Proactive

Peak’s shared library of skills, scripts, and agents responds to more than just prompts. System-wide listeners watch for relevant changes across Ridgeline’s operating environment. When a defined condition occurs, Peak assembles the necessary context and invokes the appropriate skill or agent within established permissions. Work can begin as soon as the underlying conditions change, rather than waiting for someone to notice the need and know what to ask. Peak does not just respond to employees. It can respond to the state of the business.

For example, Peak is connected to Ridgeline logs that monitor customer production environments for Ridgeline’s reporting products. When an error or anomaly occurs, Peak is triggered to investigate given specific context about the customer, the operating environment, and recent potential causes. In this role, Peak operates as an always-on partner to the engineering teams supporting our customers.

AI has made intelligence abundant in the same way the internet once made distribution abundant. Access alone was never a durable differentiator. The companies that endured were the ones that converted that abundance into systems, knowledge, and operating advantages that competitors could not easily reproduce. 

Peak allows Ridgeline to do the same by combining three company-specific assets: a shared semantic model of how Ridgeline operates, a governance layer specific to our customers and compliance requirements, and a learning loop that turns validated outcomes into reusable skills, evaluations, and automation. Because those assets remain separate from any one model, Ridgeline can continue adopting the best available intelligence without rebuilding the system around it. The models may change. The Ridgeline-specific knowledge, controls, and learning accumulated through Peak remain ours.