AI Agent Status

Applications of AI


A group of colorful plants and flowers may be wrong.
shape. AI images generated by KREA 1.

The AI ​​changes are all steadily adding more subscribers, so to those relatively new here, thank you for joining our audience.

One of the main trends I'm writing this year is the rise of AI agents, so I'd like to reorganize and share my previous article on AI agents' progress with new subscribers and longtime readers.

AI Agent Platform Shift We are refactoring how AI can be used to expand what AI can do. The development of AI inference has improved the ability of AI models to run agent AI, leading to new and innovative AI applications such as LM notebooks, deep search, AI computer use, and improvements. Generalist AI agents like Manusreleased in March and announced flooding of other improved agent AI systems.

Before we proceed, let's note that the terms AI Agent, Agent AI, AI Workflow, and AI Agent Systems all have similar, overlapping meanings for AI systems and applications.

in “Effective AI design patterns“We discussed the distinction between AI agents and AI workflows, and shared the definition of AI agents and the distinction between humanity. AI Workflow – Running AI Predefined Fixed Workflows – and AI Agents – AI systems that dynamically direct and execute the use of processes and tools.

In our terminology, An AI agent is either a system that can perform tasks dynamically and autonomously, or a fixed or dynamic workflow; An important feature of AI agents is autonomy.

Agent AI System An intelligent system that combines AI models with additional capabilities to perform tasks with minimal human monitoring. This broader term covers AI applications with agent (i.e. automation) capabilities, but it's more than an AI agent. Consider an AI assistant embedded in a word processor or coding editor. It's not the AI ​​agent itself, but it has an agent function.

One theme is the growth and improvement of the AI ​​agent ecosystem that will help AI agents. Frameworks such as Running Chain, Kuruwaiand Autogen Since 2023, we have supported the construction of AI agents and AI applications, but recent standards have helped us develop AI agents.

・Humanity released in December 2024 Model Context Protocol (MCP) has become an open industry standard To connect AI models to data and tools. It's taken over and now Thousands of MCP servers It helps to make AI models and AI agents accessible to the tools.

The latest AI models such as GPT-5, Claude 4, Claude Opus 4.1and GLM-4.5, Training of agent AI tasks, such as tool call and AI coding, is increasingly being trained.

When it comes to building AI agents, one important design element is the workflows that AI agents support. Different tasks require unique different workflows and are specific Effective AI Agent Design Patterns Covers a variety of workflows. Frameworks such as Langgraph can support predefined workflows, while AI agents can support feedback-driven, dynamic workflows.

Humanity has been developed recently A framework for developing a secure and reliable agent.

Running an AI agent involves a different way of interacting with the AI ​​chatbot. If the AI ​​chatbot answers the question, the AI ​​agent performs the task. Context Engineering is a new prompt engineering:

You need to set up whole Context – AI model memory, rapid instructions, knowledge and information support, and related tools – Help AI models complete tasks.

What do you put in the context? Provides specifications and required background information for AI agent tasks.

For example, to get production software from an AI coding tool, Drive AI coding agents with software specifications:

Vibe coding with AI speeds up software app creation, but is a technical debt recipe. Spec-driven AI coding solves that dilemmacreates well-made software in the high-speed development cycle.

AI agents now produce AI systems in 2025, from the 2023 toys to the 2024 Primitive Beta Tool. According to METR benchmarks, the length of human equivalent tasks that AI agents can do without the need for human guidance doubles every seven months.

A graph with content generated in line and dot AI may be incorrect.
Figure 2. The length of tasks that AI can do is double every seven months.

As the AI ​​agents ecosystem matured, AI agents improved and agent AI use cases expanded. The initial use of AI inference models for AI agents was very useful Deep research and AI Coding Agent. Currently, there is a recent new AI agent release, and we have added features to improve AI agents in all aspects.

I've used this phrase too much, but the progress of AI has not slowed down. The benchmark that Frontier AI models struggled with a year ago is now saturated. AI models have been significantly improved in terms of inference and tool use. The AI ​​Agent Ecosystem is much more mature than just 10 months ago, with several standardized protocols such as MCP, Agent2Agent, Agents.md, and several standardized AI Agent development, including interoperability between models and tools and standardized AI Agent development.

All these improvements in the AI ​​model and access to tools, as well as the overall AI ecosystem, have made it possible to increase the number of more advanced AI agents and agent AI applications. The resulting flood of AI agents and AI applications Surge in overall AI usageusage will increase as AI improves.

AI Agent is a new AI application layer. Agent AI has significantly changed the way software is used, replacing many other interface forms, whether it is standalone or injected with other applications. This process has just begun. Because it's good enough to make the AI ​​browser agent barely useful.

In overtime, all software is AI-first, and the software is embedded with agent AI, allowing the software to interact with the software, and in some cases, AI agents can completely displace the software and interface. There is a possibility that the creative and novel form factors of agent AI in software will still emerge.

The final line of where AI agents stand: With the improvements in the AI ​​ecosystem of AI agents and better AI models, this year's AI agents have become extremely useful in their work and personal lives. AI agents are just starting out and they say they are here.

The related questions you might ask follow-up are: How can you effectively use AI agents? What is the best AI application to use in each category? We will answer these and similar questions in subsequent articles. Please comment if you want to answer questions about a specific AI agent or cover topics.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *