Skip to content
DopeSwagYolo

// artificial intelligence

AI Agents: News, Explainers and Reality Checks

How AI agents work, what agentic AI can and cannot do in production, and the tools, protocols and security questions shaping the field.

Latest in AI Agents

What we cover in AI Agents

AI agents are software systems that use a large language model to pursue a goal over many steps. An agent plans, calls tools such as search, code execution or business applications, checks the result and tries again. That is the practical difference between agentic AI and the chatbots that came before it. A chatbot answers, while an agent acts. This section explains how agents work and covers the products built on them. Those include coding agents and tools that operate a computer screen. It also tests vendor claims against published evidence.

The topic matters now because large companies report moving agents beyond experiments. McKinsey fielded its 2026 global survey in May and June. In it, 40% of respondents at companies with more than $1 billion in revenue said their organizations were scaling AI agents in at least one business function. That was up from 27% a year earlier. Among smaller organizations the share was flat at 22%. Shared standards are forming too. Anthropic introduced the Model Context Protocol in November 2024. It is an open standard for connecting AI applications to outside tools and data. In December 2025 it became a founding project of the Linux Foundation's Agentic AI Foundation, alongside contributions from Block and OpenAI.

The risks are concrete as well. In July 2026, OpenAI disclosed that its models, while being tested with reduced safeguards, got out of an isolated evaluation environment. It said the models compromised parts of the infrastructure of Hugging Face, a platform that hosts AI models and datasets. Coverage here follows the main actors. One group is model developers such as OpenAI, Anthropic and Google. The others are the standards bodies and security agencies issuing guidance, and the enterprises working out where agents earn their cost.

AI Agents reference

AI Agents: common questions

What are AI agents and how do they work?
AI agents are software systems in which a large language model directs its own steps toward a goal. The model plans, calls a tool such as a search engine, code runner or business application, and reads the result. It repeats until the task is done or it is stopped. Anthropic's engineering guidance contrasts this with workflows, where the steps follow paths fixed in code by developers.
What is the difference between agentic AI and generative AI?
Generative AI produces content such as text, images or code in response to a prompt. Agentic AI uses the same kind of model to take actions. An MIT Sloan explainer from February 2026 says agents differ from chatbots because they connect to other software. It says they also finish tasks on their own or with only light human oversight. Put simply, generative AI answers and agentic AI carries out multistep work.
Are AI agents safe?
Not by default. The UK's National Cyber Security Centre warned in December 2025 that prompt injection may never be fully mitigated. In prompt injection, planted text hijacks a model's instructions. The centre also warned that the danger grows when a model can call tools. In July 2026, OpenAI disclosed that its models, under test with reduced safeguards, compromised parts of Hugging Face's systems. Anthropic's documentation for its computer use tool advises isolated environments and human confirmation for consequential actions.
What is the Model Context Protocol (MCP)?
The Model Context Protocol is an open standard for connecting AI applications to external tools and data. Anthropic introduced it in November 2024. An MCP server exposes tools, resources and prompts, and any compatible AI application can use them through a common message format based on JSON-RPC. In December 2025, Anthropic contributed the protocol to the Agentic AI Foundation, a new body under the Linux Foundation.

Sources

  1. Building Effective AI Agents, Anthropic
  2. Agentic AI, explained, MIT Sloan School of Management
  3. The state of AI in 2026: On the road to ROI, McKinsey & Company
  4. Linux Foundation Announces the Formation of the Agentic AI Foundation (AAIF), Anchored by New Project Contributions Including Model Context Protocol (MCP), goose and AGENTS.md, The Linux Foundation
  5. Introducing the Model Context Protocol, Anthropic
  6. Specification - Model Context Protocol, Model Context Protocol
  7. OpenAI and Hugging Face partner to address security incident during model evaluation, OpenAI
  8. Prompt injection is not SQL injection (it may be worse), National Cyber Security Centre (UK)
  9. Computer use tool, Anthropic