By Roman Aglaganov
For most of their short lives, AI chatbots have been brilliant conversationalists with no hands. They could tell you how to book a meeting, fix a bug or find a file, then leave you to actually do it. The industry’s new obsession is agents: AI systems that take actions, not just give advice.
To act, an AI needs access to the things you use: your calendar, your documents, your code, your company’s database. Until recently every one of those connections was custom-built, so every AI app needed its own adapter for every tool. Engineers call this the N×M problem. Everyone else calls it a mess of cables.
One plug for everything
In November 2024, Anthropic released the Model Context Protocol (MCP) as an open standard. The idea is simple: a tool or data source describes what it can do once, as an “MCP server”, and any AI application that speaks MCP can use it. One plug, many devices, which is why people keep calling it the USB-C port for AI.
The surprise was how quickly rivals adopted it. OpenAI added MCP support in March 2025, Google said its Gemini models would support it soon after, and Microsoft brought it to Windows and its developer tools. In December 2025, Anthropic handed the protocol to a new foundation under the Linux Foundation, so no single company controls it.
What agents can do now
Ask a modern assistant to “find last quarter’s sales figures, chart them and email the summary to the team”, and with the right connections it can plan the steps, call each tool in turn and check the results. Coding agents already open files, run tests and propose fixes on their own. It is less “chat with a robot” and more “delegate to a very fast intern”.
The catch
An agent that can act can also act wrongly, quickly. The biggest worry is prompt injection: instructions hidden in a web page or document that trick an agent into doing something its user never asked for, like forwarding private files. Sensible setups give agents the narrowest permissions possible and ask a human before anything irreversible, such as sending, paying or deleting.
The bottom line: agents are the most important shift in how people use AI since ChatGPT. Just remember that “it can do it for you” and “it should do it without asking” are two very different sentences.
