Why Agentic AI Needs Collective Human Intelligence to Deliver Contextual Insight—Keynote Insights from the 5th ICIVC 2025
Why Agentic AI Needs Collective Human Intelligence to Deliver Contextual Insight—Keynote Insights from the 5th ICIVC 2025
In a world where generative AI can synthesise language and perform reasoning at scale, one thing remains stubbornly human: context. While Agentic AI is transforming how businesses analyse, interpret, and act on data, its effectiveness continues to depend on something deceptively simple, giving it the right knowledge, in the right format, at the right time.
At the 5th International Conference on Intelligent Vision and Computing (ICIVC 2025), engineering leader Naveen Edapurath Vijayan highlighted the critical role of human insight in advancing agentic AI systems. In his keynote address, Vijayan emphasised that while generative AI can perform reasoning and language synthesis at scale, the need for context remains essentially human.
He noted that organisations today have data lakes, warehouses, and dashboards that provide access to vast amounts of information. However, he argued that these resources often fail to capture the logic, intent, and domain knowledge that drive meaningful business decisions. “What we often lack,” Vijayan said, “is a reusable layer of human insight behind the numbers. That is the context agents need if they are to be genuinely useful.”
To address this gap, he proposed a model that focuses on capturing and reusing the kinds of questions domain experts already ask. In many organisations, he explained, valuable institutional knowledge is embedded in SQL queries, dashboard filters, and ad hoc reports. This expertise is rarely documented and frequently lost over time.
The engineer outlined a framework in which SQL queries created by analysts could be collected, annotated with AI-generated explanations, and stored in a shared knowledge base. He compared the approach to “a GitHub for business logic,” where solutions to complex problems such as forecasting, churn analysis, or supply chain diagnosis could be made accessible to others in the organisation.
This model, he argued, would benefit both experts and technical users. Senior analysts would contribute depth, while newer employees could study a curated set of real-world queries and use AI-driven summaries to better understand complex logic. Over time, even AI agents trained on this repository would learn to generate not only accurate SQL but also contextually relevant insights.
The expert added that such a system should not be seen as a simple repository but as part of the agent’s extended memory, grounding its responses in organisational experience. For instance, in investigating a regional sales decline, an AI agent could analyse historical queries from past cases to suggest potential causes, recommend next steps, or generate dashboards tailored to similar scenarios.
A key point in his address was the need to preserve not only the content of queries but also the intent behind them. By storing metadata such as linked KPIs, business outcomes, and usage
context, organisations would be able to reduce duplication, improve explainability, and ensure alignment between human objectives and AI systems. “Techniques like retrieval-augmented generation and semantic pattern matching,” Vijayan noted, “can further support this process by helping systems locate and interpret the most relevant insights with greater confidence.”
The keynote concluded with a broader reflection that extended beyond AI. Vijayan stressed that progress in any domain depends not just on access to data but on how effectively human insights are captured, shared, and built upon. “When knowledge is made collective and accessible,” he said, “it enables stronger collaboration, deeper understanding, and more sustainable outcomes for both organisations and the systems they create.”
