SaaS companies may become AI harnesses

Software-as-a-service companies will increasingly compete on the systems they build around large language models, rather than on the models alone, argues Shrivu Shankar in his Substack. He writes that a company’s “harness” could become its central product capability.

Shankar uses the term broadly. A harness includes the data, tools, integrations, permissions, workflows, interfaces, and human review processes that allow a normally stateless AI model to do useful work over time. It can also include specialized AI agents for tasks such as writing specifications, generating code, reviewing work, or gathering customer feedback.

He describes a progression from employees using AI assistants individually to companies operating large numbers of background agents. In the later stages, agents take on more proactive work, while people concentrate on reviewing important decisions and applying judgment.

The model does not require a fully automated business, Shankar says. Instead, the system should direct human attention to the decisions where it adds the greatest value. An agent might collect insights from sales calls, for example, then prepare a product proposal or a set of design options for an experienced employee to assess.

Why the harness could matter

In this view, a company’s differentiation rests on how well its AI system handles four areas:

• Trust, through rules for what can be released.
• Distribution, through faster and more targeted delivery.
• Effectiveness, through tighter feedback loops.
• Domain knowledge, through the capture and use of organizational context.

Shankar expects companies to retain control of the top-level system that sets priorities and evaluates outputs. They may buy individual AI services for narrower tasks, such as turning a specification into tested code. But if an outside provider can run the full decision-making loop, he argues, the customer company may have lost much of its distinctiveness.

He also predicts that AI-native startups could gain ground where existing advantages can be converted into repeatable AI workflows. Software used in core business processes will need interfaces that other systems can control, allowing the broader AI harness to coordinate it.