@@ -12,7 +12,7 @@ Works with VS Code family of editors (Cursor, Antigravity, VSCodium, etc.).
 - APIs—_Gemini API, OpenRouter, local Ollama, etc._
 
 👉️ **Apply responses**—edits integration with checkpoints for state restoration \
-👉️ **Fully featured**—refactorings, code completions, commit messages, and more \
+👉️ **Fully featured**—code completions with reasoning models, commit messages, and more \
 🌱 **Sustainable AI coding**—designed for [prompt caching](https://platform.openai.com/docs/guides/prompt-caching) across tasks \
 ❤️ **Free and open-source**—released under the GPL-3.0 license
 
@@ -24,9 +24,9 @@ Works with VS Code family of editors (Cursor, Antigravity, VSCodium, etc.).
 
 **LLMs are brilliant, yet mindless pattern matchers.** Because they lack properties similar to mental models, they can't imagine how the correct output should look like before their context window is filled with files.
 
-Coding agents (Cursor, Claude Code, etc.) address this limitation via technique called tool calling. The model starts each task from a clean slate, by analyzing your instructions it navigates the codebase to include all necessary files in the context window. While going all-in with natural language interface for programming may sound appealing, the technology doesn't offer reliability critial in day-to-day work.
+Coding agents (Cursor, Claude Code, etc.) address this limitation via technique called tool calling. The system starts each task from a clean slate, by analyzing your instructions it navigates the codebase to include all necessary files in the context window. While going all-in with natural language interface for programming may sound appealing, these systems don't offer short feedback loops critial in day-to-day work, or cost-efficiency allowing sustainable mass adoption.
 
-Because contents of the context window determine output quality, Code Web Chat is designed in a way **the model is guided with known set of files**, enabling simpler, task focused instructions.
+Because contents of the context window determine output quality, Code Web Chat is designed in a way **the model is guided with known set of files**, enabling simple, task-focused instructions and complete edits in seconds!
 
 With files sorted by update recency, CWC helps utilize **[prompt caching](https://platform.openai.com/docs/guides/prompt-caching) across tasks**—consistently **save up to 90% on input tokens**, while spending the more expensive output tokens directly on edits you're looking for.