Kapil Chhabra

EP 75 Why Context Alone Isn't Enough for Enterprise AI Agents | WisdomAI CPO Kapil Chhabra

• Kapil Chhabra, co-founder and CPO of WisdomAI, sees the enterprise AI stack settling into layers: company data at the bottom, the models, a context layer that bridges the model’s view of the world and the company’s data, and a harness specialized for one task on top. He says the harness is little talked about today and “will become a big deal.” In his analogy, the model is the surgeon, the context is the patient history, and the harness is the hospital system with its charts, checklists, instruments and back office. Episode takeaways
• In a WisdomAI survey of senior AI, analytics and data leaders, Chhabra says 93% were using LLMs and agents to access internal data, but only 7% could say they had scaled those operations. He puts the gap down to trust: data leaders can’t yet deliver trust in the data coming out of these chat systems, and the missing piece is context, such as the company’s own definitions of ARR, churn and when the fiscal year ends, which no model knows out of the box. Episode takeaways
• A semantic layer is necessary but not sufficient, Chhabra says. It was designed for deterministic BI tools, while a chat agent reasons on the fly over any question. The context layer adds natural-language instructions (finance may mean a fiscal quarter by Q1 while marketing means a calendar quarter) and examples of what good SQL or Python looks like. He keeps institutional context, set by a person with authority such as a subject-matter expert, separate from personal memory: his own dislike of pie charts belongs in memory, not in the company’s definitions. Episode takeaways