In the context of enterprise deployments where "90% is not good enough," LLMs and truly autonomous agents are simply not quite ready to take over, as much as they hold great promise.

Still, when treated for what they are today, LLMs can serve perfectly well in many applications, perhaps not so prominent or entertaining but viable to deploy.

Data extraction from unstructured text, summarization, and classification are examples close to our business, focusing on the niche of market research automation.

These can work well once integrated into a human-expert-engineered, customer-specific workflow (including 'traditional' automation) and today's company operations.

The diagram below illustrates a simplified workflow in our particular domain. At Robotic Online Intelligence (ROI), we have been deploying LLMs and 'AI Workflow' structures within such a broader framework.