Why agentic AI?
Where autonomous AI agents meet healthcare, science and entrepreneurship.
AI systems are moving from answering questions to taking actions: planning multi-step tasks, using tools, collaborating with people and with other agents, and learning from feedback. The shift opens real opportunities and raises hard questions about reliability, safety and accountability.
The Agentic AI Workshop 2026 brings researchers, practitioners, founders and investors to IIT Delhi to take stock of where agentic AI stands, where it is heading, and how it can create real impact in healthcare, scientific discovery and enterprice.
Agentic AI describes systems that pursue goals with a degree of autonomy. Rather than producing a single response, an agent runs a loop, usually with a large language model as its reasoning engine, combined with memory, tool use and human oversight.
What is an agentic system?
Plan, delegate, verify, repeat
An agentic system brings together specialized AI agents that collaborate, exchange information, and iteratively refine their outputs to accomplish complex tasks.
- orchestratorno source exists, so instruct drop
- reasonerdrops the unsupported claim
- verifierall 19 claims match their source
- orchestratorreturns answer, citations and trace
Two failed attempts, one success. Figures are illustrative, at roughly today's frontier-model rates. The retries are the cost, and the verifier is what you are buying.
An orchestrator coordinates specialised agents for retrieval, reasoning, and verification through shared memory. If verification fails, it re-plans instead of repeating the same attempt. However, reliability, evaluation, tool safety, and human oversight are essential for deploying trustworthy agentic systems.