# Micheal Lanham - AI Agents in Action (Highlights)
## Metadata
**Review**:: [readwise.io](https://readwise.io/bookreview/56990768)
**Source**:: #from/readwise #from/zotero
**Zettel**:: #zettel/fleeting
**Status**:: #x
**Authors**:: [[Micheal Lanham]]
**Full Title**:: AI Agents in Action
**Category**:: #books #readwise/books
**Category Icon**:: 📚
**Highlighted**:: [[2025-12-21]]
**Created**:: [[2025-12-27]]
## Highlights
- Finally, fine-tuning is a feature added to models that refines the input data and model training to better match a particular use case or domain ([Page 47](zotero://open-pdf/library/items/EVUUX6GA?page=46&annotation=ERWELLXX)) ^968473792
- Assistant role —Can be used to capture the message history of previous responses from the LLM. It can also inject a message history when perhaps none existed. ([Page 52](zotero://open-pdf/library/items/EVUUX6GA?page=51&annotation=2MVAY5C3)) ^968473793
- LM Studio is a free download that supports downloading and hosting LLMs and other models locally for Windows, Mac, and Linux. ([Page 55](zotero://open-pdf/library/items/EVUUX6GA?page=54&annotation=LUWYAITT)) ^968473794
- An appealing feature of LM Studio is its ability to analyze your hardware and align it with the requirements of a given model. ([Page 56](zotero://open-pdf/library/items/EVUUX6GA?page=55&annotation=SIBKDDGS)) ^968473795
- Rubber ducking is a problem-solving technique in which a person explains a problem to an inanimate object, like a rubber duck, to understand or find a solution. ([Page 72](zotero://open-pdf/library/items/EVUUX6GA?page=71&annotation=8TUQ978H)) ^968473796
- Injects the sample output as the “previous” assistant reply ([Page 75](zotero://open-pdf/library/items/EVUUX6GA?page=74&annotation=WQDBBCR5)) ^968473797
- We’ll explore a project from Microsoft called AutoGen, which supports multiple agents but also provides a studio to ease you into working with agents. ([Page 127](zotero://open-pdf/library/items/EVUUX6GA?page=126&annotation=N6LV8BLB)) ^968473798
- Then, we’ll transition to CrewAI, a self-proposed enterprise agentic system that takes a different approach. CrewAI balances role-based and autonomous agents that can be sequentially or hierarchically flexible task management systems. ([Page 127](zotero://open-pdf/library/items/EVUUX6GA?page=126&annotation=CNRRV94K)) ^968473799
- Test Changes Systematically is such a core facet of prompt engineering that Microsoft developed a tool around this strategy called prompt flow ([Page 345](zotero://open-pdf/library/items/EVUUX6GA?page=344&annotation=FLBWHCQM)) ^968473800
- This time, enter the following prompt and include the word system in lowercase, followed by a new line ([Page 349](zotero://open-pdf/library/items/EVUUX6GA?page=348&annotation=YCY89SJM)) ^968473801
- At a basic level, an agent profile is a set of prompts describing the agent. It may include other external elements related to actions/tools, knowledge, memory, reasoning, evaluation, planning, and feedback. The combination of these elements comprises an entire agent prompt profile. ([Page 355](zotero://open-pdf/library/items/EVUUX6GA?page=354&annotation=QDIIHWBH)) ^968473802
- Prompt flow, VS Code extension —Refer to appendix A for details on installing extensions. ([Page 356](zotero://open-pdf/library/items/EVUUX6GA?page=355&annotation=7GZVXCW3)) ^968473803
- Grounding is a concept that can be applied to profile and prompt evaluation—it defines how well a response is aligned with a given rubric’s specific criteria and standards. You can also think of grounding as the baseline expectation of a prompt or profile output ([Page 373](zotero://open-pdf/library/items/EVUUX6GA?page=372&annotation=LSYAD5LA)) ^968473804
- This section will employ another LLM prompt/profile for evaluation and grounding. ([Page 375](zotero://open-pdf/library/items/EVUUX6GA?page=374&annotation=5L6NBLYQ)) ^968473805