decomposition (planing): break down the task into achievable goals. Searches should also be subdivided

recomposition: Sanatize the previous work to be ready for summary and other agents

validation: proper ways to verify that steps are in track, otherwise the afterwork is wasted

topologies: - State Machine - not all is groupchat - GPT4 needs less agents

self-refine: Here you can find a code example for a self-improvement flow.

videos:
- **Caching:** Caching is a technique for improving the performance of Autogen applications. By caching the results of past queries, Autogen can avoid having to recompute these results, which can save time and resources.
- **Best practices:** There are a number of best practices that developers can follow to create high-quality Autogen applications. These best practices include using clear and concise code, using descriptive variable names, and testing Autogen applications thoroughly.
- **Enhanced inference:** Enhanced inference is a technique for improving the accuracy of Autogen applications. Enhanced inference techniques can be used to identify and correct errors in the output of Autogen models.
- **Multi-agent setup:** Multi-agent setup is a technique for using multiple Autogen agents to solve a single problem. Multi-agent setup can be used to create more powerful and sophisticated Autogen applications.
- **LLM daisy-chaining:** LLM daisy-chaining is a technique for connecting multiple Autogen models together. LLM daisy-chaining can be used to create even more powerful Autogen applications.
- **Teachable Agent:** A Teachable Agent is a new type of Autogen agent that can learn from its own experiences. Teachable Agents have the potential to become even more powerful and versatile than traditional Autogen agents.
- **Function calling:** Function calling is a technique for calling functions from within Autogen code. Function calling can be used to create more complex and modular Autogen applications.
- **Templating:** Templating is a technique for generating text from templates. Templating can be used to create Autogen applications that can generate different types of content, such as emails, letters, and reports.
  • video video in agentic design patterns gpt3 with self reflection surpass gpt at zero-shot