What is agentic Artificial Intelligence (AI) ? Explain its working. Describe its applications with suitable examples. Discuss the advantages, risks and challenges associated with agentic AI systems.
15 marks · answer in 250 words · directive: Explain / Describe / Discuss
The core of a good answer
Agentic AI refers to systems, usually built on large language models, that pursue a goal on their own by planning steps, calling tools and software, observing results and adjusting; they promise large productivity gains, but autonomy brings new risks such as hijacking through prompt injection, errors that compound across actions and unclear accountability, so they need guardrails and human oversight.
An outline written by upsc.help, not an official UPSC answer; every fact checked by a separate review. How we work
What the question asks
Four parts: definition, working (the perceive, plan, act and learn loop), applications with examples, and a balanced discussion of advantages, risks and challenges. Use concrete, dated examples rather than generic statements.
Answer outline
Introduction
Unlike a chatbot that answers one prompt, agentic AI acts: given a goal, it breaks it into steps and uses tools such as browsers, code and databases to complete the task with limited human input.
Working
- Reasoning core: a large language model interprets the goal and plans sub-tasks.
- Tools: the agent calls APIs, searches, runs code or operates a computer; standards such as the Model Context Protocol (introduced by Anthropic in November 2024) connect models to external tools and data.
- Memory and feedback: it stores context, checks results and replans (a loop of perceive, plan, act, reflect).
- Multi-agent systems split work among specialised agents; human-in-the-loop checkpoints approve high-impact actions.
Applications with examples
- Computer and browser use: Anthropic’s Claude computer use (October 2024) and OpenAI’s Operator (January 2025) and ChatGPT agent (July 2025) fill forms, book and order online.
- Software engineering: coding agents write, test and fix code across a repository.
- Business and government: customer service, procurement, document processing, grievance handling.
- Science and health: literature review, lab automation, triage support.
- Agriculture and citizens: multilingual advisory agents on top of India’s digital public infrastructure.
Advantages
- Productivity: automates multi-step knowledge work, round the clock.
- Scale and inclusion: can deliver services in local languages at low cost.
- Speed in research, cyber defence and operations.
Risks
- Agent hijacking: malicious instructions hidden in emails, files or web pages redirect the agent; NIST’s CAISI found attack success on one agent rose from 11% to 81% with new red-team attacks (January 2025).
- Compounding errors and hallucinations across many steps; unintended actions with money or data.
- Privacy and data leakage, since agents need broad access; misuse for fraud, cyber attacks and disinformation.
- Job displacement in routine white-collar work.
Challenges and governance
- Accountability and liability when an autonomous agent causes harm.
- Testing and evaluation are hard because behaviour is probabilistic; repeated attempts raise attack success (CAISI: 57% to 80% over 25 attempts).
- Compute, data and talent needs; India’s IndiaAI Mission (March 2024, Rs 10,371.92 crore) addresses compute and datasets.
- India AI Governance Guidelines (5 November 2025) take a risk-based, principle-based approach using existing laws such as the IT Act and DPDP Act, without a new AI law for now.
Conclusion
Agentic AI moves AI from advice to action. Its benefits are real, but so is the risk surface; least-privilege access, human approval for consequential actions, audit trails and testing standards should grow with its adoption.
Examples and data to cite
- Model Context Protocol introduced by Anthropic on 25 November 2024; donated to the Agentic AI Foundation under the Linux Foundation in December 2025.
- OpenAI Operator released January 2025; ChatGPT agent released July 2025.
- NIST CAISI technical blog on agent hijacking evaluations, 17 January 2025.
- IndiaAI Mission approved March 2024 with Rs 10,371.92 crore over five years; 38,000 GPUs made available (PIB, December 2025).
- India AI Governance Guidelines released 5 November 2025 (PIB, 19 December 2025).
Facts to remember
- An AI agent is goal directed and acts through tools; a generative model alone only produces content.
- Indirect prompt injection is the main security weakness because trusted instructions and untrusted data share one input.
- India hosted the AI Impact Summit in February 2026.
- India’s guidelines rely on sectoral regulators and existing law (IT Act, DPDP Act).
Where it sits in the syllabus
- GS3: IT, space, biotech, IPR
Awareness in the fields of IT, Space, Computers, robotics, nano-technology, bio-technology and issues relating to intellectual property rights.
- GS3: Science and technology in everyday life
Science and Technology: developments and their applications and effects in everyday life.