Tech Desk: India’s inclusion in Anthropic’s Project Glasswing, granting access to the advanced Claude Mythos Preview, marks a pivotal moment for the nation’s cybersecurity and AI landscape. Designed for sophisticated vulnerability detection and threat response, Mythos promises to elevate India’s digital defenses, shifting security from a reactive stance to a proactive one, as experts like Srinivas Padmanabhuni note.

While the technological leap is significant, the long-term debate surrounding this agentic AI model pivots sharply to its “AI economics.” CloudSEK founder Rahul Sasi highlights that the true recurring cost isn’t training the AI, but rather ‘inference’—the continuous process of hosting the model, making it reason, call tools, and retry until a useful outcome is achieved. This makes ‘token cost’ a critical factor.

Application-layer companies, rather than foundation model developers, will bear the brunt of optimizing these recurring expenses. They’ll need to innovate in context management, retrieval, caching, and workflow orchestration to keep costs down. Viral Shah of JuliaHub adds that India’s long-term AI competitiveness hinges on robust infrastructure like energy, data centers, and silicon, areas where government initiatives can play a crucial role.

Ultimately, the question isn’t just about AI agents’ capabilities, but how much “token burn” and associated dollar cost it takes to achieve a result. Sasi warns that in some complex security operations, an AI agent’s persistent reasoning and tool-calling might even outprice a human analyst with established business context. The economics of deploying AI at scale will define its true impact.