What is agentic AI today, and what do we want it to be?
The article examines the current state of agentic AI—systems that can autonomously pursue goals—and explores what capabilities and safeguards society should aim for as the technology continues to evolve.
Background
MIT News piece surveying the current state and open questions around "agentic AI" — AI systems that can take autonomous, multi-step actions rather than just responding to prompts. The article draws on MIT researchers (e.g. from CSAIL, the MIT-IBM Watson AI Lab) to distinguish between today's limited "agents" (often just LLMs wrapped with tool-use capabilities, prone to errors and hallucinations) and a longer-term vision of systems that can plan, reason, adapt, and act reliably in the world. Key context: the AI industry is racing to build and deploy agents (e.g. coding agents, browser-controlling bots, customer-service automation), but fundamental problems remain — unreliable reasoning, lack of robust world models, security risks, and unclear frameworks for accountability. The piece isn't reporting a specific breakthrough; it's a reflective essay on the gap between the marketing hype and the research reality, and what technical and societal design choices need to be made to steer agentic AI toward beneficial outcomes.