41 lines
3.4 KiB
Markdown
41 lines
3.4 KiB
Markdown
## Summary
|
||
The article discusses the advancements in AI agent frameworks, specifically LangChain and LangGraph. LangChain is a user-friendly tool for building AI agents, while LangGraph is a lower-level framework for highly custom and controllable agents. The article also highlights the importance of agentic workflows, which allow AI systems to make decisions and adapt based on their environment. Five essential agentic workflow patterns are identified: reflection, tool use, planning, learning, and optimization.
|
||
- ## Mermaid Mindmap
|
||
- {{renderer :mermaid_694c8c42-5621-4d95-bd16-8261a5eb4f1c, 3}}
|
||
- ```mermaid
|
||
graph TD
|
||
A[LangChain] --> B[User-Friendly AI Agent Framework]
|
||
A --> C[Standard Tool Calling Architecture]
|
||
A --> D[Provider Agnostic Design]
|
||
A --> E[Middleware for Customization]
|
||
E --> F[Note: MOSAICO]
|
||
E --> G[LangGraph: Lower Level Framework]
|
||
G --> H[Useful for Highly Custom and Controllable Agents]
|
||
G --> I[Support Production-Grade, Long Running Agents]
|
||
B --> J[Background Color: Green]
|
||
C --> J
|
||
D --> J
|
||
E --> J
|
||
F --> J
|
||
H --> J
|
||
A --> K[Top AI Agentic Workflow Patterns]
|
||
K --> L[AI Automation]
|
||
K --> M[Agentic Workflows]
|
||
K --> N[Reflective Systems]
|
||
K --> O[Tool Integration]
|
||
L --> P[An agentic workflow doesn’t just respond to a single instruction. Instead, it operates with a degree of autonomy, making decisions about how to approach a task, what steps to take, and how to adapt based on what it discovers along the way.]
|
||
L --> Q[An agentic system might first search the web for current information on the topic, then organize the findings into themes, draft sections of the report, review each section for accuracy and coherence, revise weak areas, and finally compile everything into a polished document.]
|
||
L --> R[Instead of generating output in a single pass, agentic workflows involve cycles where the agent takes an action, observes the result, and uses that observation to inform the next action]
|
||
L --> S[Agentic workflows bring this same adaptive, iterative quality to AI systems.]
|
||
K --> T[Five Essential Agentic Workflow Patterns]
|
||
T --> U[Reflection Pattern: Works best for tasks where quality matters more than speed and where there are subjective aspects that benefit from review.]
|
||
T --> V[Tool Use Pattern: Agents are equipped with a set of capabilities they can invoke when needed. These might include web search engines, APIs, code interpreters, database query tools, and file system access.]
|
||
K --> W[The Fundamental Challenge: Context windows are constrained not by raw token capacity but by attention mechanics]
|
||
W --> X[U-Shaped Attention Curves, and Attention Scarcity]
|
||
W --> Y[Effective Context Engineering Means Finding the Smallest Possible Set of High-Signal Tokens that Maximize the Likelihood of Desired Outcomes.]
|
||
K --> Z[7 Tiny AI Models for Raspberry Pi - KDnuggets]
|
||
Z --> AA[Aggressive Quantization]
|
||
Z --> AB[llama.cpp: What's llama.cpp? #TODO To be checked.]
|
||
Z --> AC[Tool Calling, Vision Understanding, and Structured Outputs]
|
||
Z --> AD[Vision-Language Model: Merges Computer Vision and Natural Language Processing, Allowing Them to Understand, Interpret, and Generate Content from Both Images/Videos and Text Inputs, Enabling Tasks Like Describing Photos (Captioning), Answering Questions About Visuals (VQA), Generating Images from Text, and Understanding Complex Documents.]
|
||
``` |