Shrivu Shankar argues that SaaS companies will become operating frameworks that provide models with context, tools, permissions and review procedures. He forecasts a shift from people invoking agents to background agents proposing work and requesting human judgment when needed. Companies should own the higher-level structure that decides what to build and reviews outcomes, while potentially buying individual tools. He also acknowledges current model reliability limits. Anthropic’s examples of session logs and separation of execution environments provide technical background for the recoverability of such systems.