What exactly are AI Agents by Aki Ranin (Earthshots Collect

What exactly are AI Agents?
Aki Ranin, Head of AI at Earthshots Collective | Deep Tech & AI Investor | 2x Founder | Published Author

apidays Singapore 2025
Where APIs Meet AI: Building Tomorrow's Intelligent Ecosystems
Marina Bay Sands Expo & Convention Centre
April 15 & 16, 2025

------

Check out our conferences at https://www.apidays.global/

Do you want to sponsor or talk at one of our conferences?
https://apidays.typeform.com/to/ILJeAaV8

Learn more on APIscene, the global media made by the community for the community:
https://www.apiscene.io

Explore the API ecosystem with the API Landscape:
https://apilandscape.apiscene.io/

Transcript

1. What exactly are AI Agents?
2. Timeline of AI development 1950 1970 1990 2010 2030 Artificial Intelligence Neural Networks Machine Learning Deep Learning GenAI Agents
3. Artificial Intelligence: Definitions and history “I propose to consider the question, can machines think?” – Alan Turing, 1950 Turing Test, 1950 IBM Deep Blue, 1996 AlphaGO, 2016 ChatGPT, 2023 ARTIFICIAL INTELLIGENCE
4. Agents - making LLMs do tasks and jobs AI Agency Human Agency Agency Autonomy Prompt Agents perform complex queries and tasks as directed by the human Workflow Agents perform tasks as part of a workflow with limited autonomy Worker Agents take on human-like roles and perform tasks on their own Cursor, Operator, Deep Research Runner H, Gumloop, Agentforce Devin, 11x, Manus GENERATIVE AI - AGENTS
5. Agents - what’s under the hood GENERATIVE AI - AGENTS
6. Agents - what’s under the hood GENERATIVE AI - AGENTS SCAFFOLDING • Memory • Chain-of-thought • Reasoning • Tool access ◦ Browser ◦ Software ◦ Computer ◦ External APIs
7. Model Context Protocol (MCP) – “USB-C for LLMs” GENERATIVE AI - AGENTS Anthropic’s open-source project is quickly becoming an industry standard for API and LLM interoperability. • Inspired by Language Server Protocol (IDEs) and based on JSON-RPC 2.0 • SDKs for Python, TypeScript, Java, Kotlin, and C# • Servers provide: ◦ Authorization (OAuth 2.1) ◦ Resources (context/data) ◦ Prompts (templates and workflows) ◦ Tools (functions to call) ◦ Sampling (agentic behaviors)
8. Prompt Agents - You ask the AI to complete a task in a chat window GENERATIVE AI - AGENTS Feels just like a chatbot, except that it uses many tools and takes minutes instead of seconds to give you results. Tools: web search, web browser, 3rd party APIs, code execution, command line, etc. Examples: Github Copilot, Cursor, Claude Computer Use, OpenAI Operator, as well as Deep Research in Google, OpenAI, Perplexity, and xAI flavors.
9. Workflow Agents - The AI completes tasks semi-autonomously as part of a pre-determined workflow GENERATIVE AI - AGENTS No chat is required, can be triggered by user or other systems. Workflows use AI to carry out specific tasks or make decisions. Triggers: email, database, file, or API Tools: databases, knowledge bases, cloud storage, 3rd party APIs Examples: Stack.ai, Gumloop, Agentforce, Runner H
10. Worker Agents - The AI autonomously carries out tasks as part of a role description GENERATIVE AI - AGENTS Feels more like a new employee. You engage via normal tools and channels you would with any employee (except f2f). Pricing not per user but per agent. Tools: email, instant messaging, task managers, code repositories, etc. Examples: Devin, 11x, Claude Code
11. Agent2Agent Protocol (A2A) – Towards an Internet of Agents GENERATIVE AI - AGENTS Google has just announced their own protocol for agent-to-agent communication. We can think of A2A as TCP for AI. A way to connect Agents to each other, the internet, and the digital economy, without humans to slow things down. Running on crypto rails or not? • Stripe already launched MCP server • Acquired Bridge (stable coins as API)