AI Agents Video Review

https://www.youtube.com/watch?v=FwOTs4UxQS4

ChatGPT, Gemini, Claude are applications built on top of an LLM.
When is my next coffee chat - ChatGPT fails as it doesn’t have access to my calendar.
- Have limited knowledge of proprietary informatoin - personal information or internal company data.
- LLMs are passive - they wait for prompts and then respond.
Have the LLM check my calendar before the response is provided.
- What if the follow-up question is, what will the weather be like that day - the LLM can’t compute this, as there is nothing in the Calendar that says about weather.
The above is an AI Workflow.
AI has pre-defined paths. PATH = CONTROL LOGIC.
What about adding an API for weather to the LLM as well?
- If a human is the decision maker, there is no AI involvement.
- RAG is a type of AI workflow.
He uses GOogle Sheets –> Perplexity –> Claude.
- Reason - compile links, summarise articles, write posts.
- Action - Sheets, Perplexity and then Claude for writing the posts.
However, ultimately there is a human decision maker. That should be replaced by an AI agent.
- This should reason - What’s the most efficient way to compile news articles?
ReAct Framework - All AI Agents must Reason and then Act.
AI Agent also need the ability to iterate.
With a human decision maker, the find post that is written needs to be iterated a few times to make something good. Goes over multiple forms of Iteration.
- An AI agent can do the same thing autonomously.
  - The AI agent would add another LLM to critique its own output.