Q&A: What is agentic AI today, and what do we want it to be?

2026-06-30 · Source: MIT News - Artificial intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Intermediate, medium

Summary

Agentic AI, defined as AI that takes actions in the world—physical or digital—differs from generative AI, which primarily creates content. A November 2025 report by MIT Sloan School of Management and Boston Consulting Group indicated 35 percent of surveyed businesses had deployed AI agents, with another 44 percent planning implementation soon. These systems typically use a foundational generative AI model, such as Claude, augmented with application-specific tools like calculators or data access. A significant challenge in developing agentic AI is the scarcity of training data for complex tasks, often necessitating trial-and-error learning in dynamic environments. Promising applications include coding agents, which leverage feedback loops for problem-solving. However, risks involve human oversight failures leading to bugs or data leaks, and potential de-skilling as reliance on agents grows. The future of agentic AI may involve new architectures beyond text-trained language models to handle diverse data modalities like video and physical forces, or advanced LLMs integrated with sensors and actuators.

Key takeaway

For AI Engineers developing agentic systems, recognize that current models, while powerful, require careful integration of application-specific tools and robust training strategies. You must prioritize rigorous verification of agent outputs to mitigate risks like data leaks and bugs, especially given the ease of deployment. Implement human-in-the-loop processes for high-stakes applications to prevent errors from vague instructions and to counter potential de-skilling effects.

Key insights

Agentic AI extends generative models with tools to perform actions, facing data scarcity and human oversight risks.

Principles

• Agentic AI acts; generative AI creates.
• Foundation models are core to agents.
• Balance automation with human assistance.

Method

Agentic AI systems start with a generative AI core, then integrate application-specific tools. Training often involves trial-and-error in dynamic environments due to data scarcity.

In practice

• Use coding agents for iterative problem-solving.
• Implement verification steps for agent outputs.
• Consider human-in-the-loop for high-stakes tasks.

Topics

• Agentic AI
• Generative AI
• AI Agents
• Large Language Models
• AI Development
• AI Risk Management

Best for: AI Scientist, AI Engineer, Director of AI/ML

Related on AIssential

• 4 years of observing and living through the AI era — Generative and agentic AI tools offer unprecedented automation and creation capabilities, shifting the focus to human validation and decision-making.
• 15+ Solved Agentic AI Projects with Github Links — Agentic AI systems move beyond generative models to reason, plan, and execute tasks autonomously.
• 🧠 Sam Altman says OpenAI’s models are “beginning to find critical vulnerabilities” — AI models are advancing rapidly across security, coding, and voice synthesis, demanding new approaches to safety and efficiency.
• AI Agents Explained Like You’re an Engineer: A Beginner’s Guide to How Modern AI Actually Works — Modern AI agents operate as complex software systems, integrating multiple steps into a continuous decision-making loop.
• A Language For Agents — New programming languages designed for AI agents will emerge as coding costs decrease and agentic engineering grows.
• Your Agent Loop Is Fine. Your Tools Are Why It Breaks. — Agent reliability hinges on well-defined tools, not complex orchestration loops.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by MIT News - Artificial intelligence.