Enterprise hits and misses - agentic AI project failure versus success, open source versus AI, and the perils of disconnected CX
The article discusses the challenges and successes of implementing agentic AI in enterprises, emphasizing the importance of context at inference, constraining large language models within compound systems, and distinguishing between off-the-shelf and domain-specific models. It highlights that AI success requires careful tool selection, use case design, and organizational readiness, advocating for modest, repeatable AI successes rather than overambitious goals.
For enterprises delving into agentic AI projects, the key takeaway is to adopt a risk-managed mentality that prioritizes "context at the time of inference" and "constraining LLMs within a compound systems architecture." This involves integrating domain-specific models with other machine learning forms and deterministic systems to enhance reliability and governance. Additionally, building AI readiness by dismantling data silos and fostering cross-departmental collaboration for process rethinking and governance is crucial, ensuring that AI initiatives are sustainable and aligned with business outcomes.