Updated reasoning loop controlling description.

_posts/2024-04-19-Data-Agents-With-LalaIndex.md

@@ -20,7 +20,7 @@ The main core steps of a Data Agent include:
    - [**Function Calling Agent:**](https://docs.llamaindex.ai/en/stable/examples/Agent/openai_Agent_parallel_function_calling/) A unified abstraction that utilizes function calling capabilities of different LLMs to call given tools.
    - [**Chain-of-Abstraction Agent:**](https://docs.llamaindex.ai/en/stable/examples/agent/coa_agent/) Implements a generalized version of the strategy described in the [original paper](https://arxiv.org/pdf/2401.17464.pdf). It enables LLMs to learn more general reasoning strategies that are robust to shifts of domain knowledge relevant to different reasoning questions.
    - [**Retrieval Augmented:**](https://docs.llamaindex.ai/en/stable/examples/Agent/openai_Agent_retrieval/) Utilizes an Agent together with a tool retriever to manage an index on an arbitrary number of data tools, reducing latency and cost by retrieving only relevant tools.
-   - [**Controlling Reasoning Loop:**](https://docs.llamaindex.ai/en/stable/examples/Agent/return_direct_Agent/?h=return_direct) Allows modification of the Agent reasoning loop, where the output is returned directly instead of using an LLM, useful for speeding up response times.
+   - [**Controlling Reasoning Loop:**](https://docs.llamaindex.ai/en/stable/examples/Agent/return_direct_Agent/?h=return_direct) Allows modification of the Agent reasoning loop, where the output is returned directly instead of using an LLM, useful for speeding up response times. This also allows direct output returns, reducing costs and enhancing response efficiency.
    - [**Step-wise Controllable:**](https://docs.llamaindex.ai/en/stable/examples/agent/agent_runner/agent_runner/) Provides more granular control of the Agent by separating task creation and execution, enabling sharing feedback to the Agent as it completes tasks.
 
 - **Tool Abstractions:** Based on the decision-making process in the Reasoning Loop, the Agent selects the most relevant tools to fetch or modify data. Tool abstractions provide a structured way to define how Data Agents interact with data or services Types of Tools include: