Pablo Sanzo / AI agents

Definitions

"From AI model perspective: AI model which can perform actions via tools"

1. Agents Course (Hugging Face, Feb-2025)
   • "An AI model capable of reasoning, planning, and interacting with its environment"
2. AI Engineer Conference (Anthropic, Feb-2025)
   • "Agents are models using tools in a loop"

"From software perspective: software where AI model manages the control logic"

1. Building effective agents (Anthropic, Dec-2024)
   • "Workflows are systems where LLMs and tools are orchestrated through predefined code paths"
   • "Agents are systems where LLMs dynamically direct their own processes and tool usage"
   • "Workflows & Agents = Agentic systems"
2. Introduction to Operator & Agents (OpenAI, Jan-2025)
   • "AI agents are AI systems that can do work for you independently; you just give them a task"
3. What are AI agents? (IBM, Jul-2024)
   • "AI agents are compound AI systems where an LLM is in charge of the control logic"
4. Introduction to Agents (Hugging Face, Jan-2025)
   • "AI agents are programs where LLM outputs control the workflow"
   • "Spectrum as you give more or less power to the LLM on your workflow"
5. What is an AI agent? (LangChain, Jun-2024)
   • "An AI agent is a system that uses an LLM to decide the control flow of an application"
6. Agents Masterclass (LangChain, May-2023)
   • "Use LLM as the "reasoning engine"
   • "Non-deterministic sequence of actions"
7. Defining Agents (OpenAI, Feb-2025)
   • "An agent is an AI application consisting of a model equipped with instructions that guide its behavior, access to tools that extend its capabilities, encapsulated in a runtime with a dynamic lifecycle"
   • "Agent = Model + Instructions + Tools + Runtime"
8. Barista Bot (Replit, Jul-2024)
   • "Agents are applications that use large language models (LLMs) to complete complex tasks"
   • "Agent as a smart assistant where the LLM acts as the brain, directing various operations to meet a user's request"
   • "Agents combine the thinking power of an LLM with specialized tools to get jobs done"

Bonus: Steve Jobs (1984)

"Well, the types of computers we have today are tools. They're responders: you ask a computer to do something and it will do it. The next stage is going to be computers as "agents." In other words, it will be as if there's a little person inside that box who starts to anticipate what you want. Rather than help you, it will start to guide you through large amounts of information. It will almost be like you have a little friend inside that box. I think the computer as an agent will start to mature in the late '80s, early '90s."

Examples

1. Coding agent:
• Cursor
• Windsurf
• Cline
• Aider
• Goose (Block)
• Trae (ByteDance)
• SWE agent from OpenAI in Q1-Q2 2025?
2. Web operator:
• Operator (OpenAI) (Vision) (Computer-Using Agent, CUA)
• Open Operator (Browserbase) (Stagehand from Playwright)
• Browser Use Cloud (Browser Use) (Vision + Helium from Selenium)
• Anthropic Computer Use Demo (Replit)
3. Computer use:
• Anthropic Computer Use
• Agent.exe
• UI-Tars (Bytedance)
4. Deep research:
• Deep Research (OpenAI)
• Deep Research (Google)
• Deep Research (Perplexity)
• Storm (Stanford)
• Open Deep Research (Hugging Face, Smolagents)
• Open Deep Research (Firecrawl)
• Competitive Analysis Workflow (IBM, BeeAI)
5. Sales:
• Sales Associate agent from OpenAI imminent?
6. General:
• Manus

Classifications

1. By STRUCTURE: Single agent (logic in instructions) vs. Multi-agent (e.g. logic as graph)
2. By how they REASON: in Language vs. in Code (e.g. Smolagents CodeAgent)
3. By how they are TRIGGERED: Reactive (when called) vs. Proactive (e.g. LangGraph Ambient Agents)
4. By INTERACTION MODEL: One-off vs. Iterative (Chat)
5. By their MARKET: B2B (Enterprise; SMB) vs. B2C (Consumer)
6. By their TARGET: Vertical (one function e.g. Cursor) vs. Horizontal (across functions e.g. Ottogrid)
• Open to debate where to draw the line