# Cursor 2.0 Is INSANE: Multi-Agents Build Deploy Apps AUTOMATICALLY!

Source: https://www.youtube.com/watch?v=m6cMChpRIgM
Recap page: https://rapidrecap.app/video/m6cMChpRIgM
Generated: 2025-11-15T19:04:13.022+00:00

---
## Quick Overview

Cursor 2.0 features an advanced multi-agent system, demonstrated by successfully building, testing, deploying an AI-powered bank statement analyzer application using Terraform on Google Kubernetes Engine (GKE), and optimizing Kubernetes resources, all orchestrated through agent commands within the IDE.

**Key Points:**
- Cursor 2.0 introduces an integrated multi-agent system, allowing parallel execution of up to four agents (Composer 1, Sonnet 4.5, GPT-5.1 Fast, Grok 4 Fast) on a single task (00:03, 13:48).
- An AI-powered Bank Statement Analyzer application was developed, accepting Excel, PDF, and DOCX files, which required setting up backend/frontend/database components (2:24).
- The application deployment involved automating Terraform to create a GKE cluster and deploying the service to Cloud Run, followed by migrating the deployment to Kubernetes (1:04, 1:11, 1:23, 7:43).
- The agent successfully performed complex infrastructure tasks, including deleting the Cloud Run service and then using Terraform to deploy to Kubernetes, all in under 10 minutes (7:57, 8:53).
- Kubernetes resource optimization was demonstrated by analyzing pod usage (CPU hitting 94.1% limit) and suggesting/implementing fixes like adjusting HPA targets and increasing CPU limits, completed in about one minute (11:54, 12:55).
- The multi-agent system allows for selecting specific models (like Composer 1, Sonnet 4.5, GPT-5.1 Fast, Grok 4 Fast) and scaling the agent count up to 4x for parallel work (13:32, 13:48).
- The agent successfully implemented a login feature for the application, handling JWT creation, verification, and local storage of tokens (14:17).

![Screenshot at 0:00: The initial Cursor 2.0 interface showing four active agents \(Composer 1, GPT-5.1 Fast, Gemini 2.5 Flash, Grok 4 Fast\) concurrently working on the task 'Implement login feature for application'.](https://ss.rapidrecap.app/screens/m6cMChpRIgM/00-00-00.png)

**Context:** This video showcases the advanced capabilities of Cursor 2.0, focusing heavily on its multi-agent system and its ability to handle complex, multi-step software development and DevOps tasks automatically. The presenter demonstrates building a full-stack 'Bank Statement AI Analyzer' application, testing it, deploying it first to Google Cloud Run, and then migrating and optimizing it on a Google Kubernetes Engine (GKE) cluster, all driven by natural language prompts to the agents.

## Detailed Analysis

The video demonstrates Cursor 2.0's powerful multi-agent capabilities by walking through the end-to-end development and deployment of a complex application: an AI-powered Bank Statement Analyzer. First, the user prompts the agents to create the application, which involves setting up frontend, backend, and database components (2:24). The agent successfully implements JWT-based login functionality (14:17). The agent then demonstrates infrastructure automation by deploying the application first to Google Cloud Run (1:11) and then automating the destruction of the Cloud Run service and creation of a GKE cluster using Terraform scripts (7:43, 8:00). Following deployment to Kubernetes, the agent analyzes resource usage, identifying high CPU utilization (up to 94.1% of the limit) and suggests specific optimizations, which it then implements by patching the deployment YAML configurations, all within minutes (11:54, 12:24). The video highlights the ability to select and scale multiple LLMs (Composer 1, Sonnet 4.5, GPT-5.1 Fast, Grok 4 Fast) concurrently for parallel task execution (13:32). Finally, the agent successfully destroys all created resources using `terraform destroy` (10:34). The entire workflow showcases an advanced, highly automated AI development environment.

### Cursor 2.0 Features

- Composer is a frontier model 4x faster than similar models
- Multi-agent interface allows focusing on outcomes rather than files
- TestSprite integration for automated testing and reporting (5:56)

### Bank Statement Analyzer Development

- Application accepts Excel, XLSX, .aix, PDF, Word, .docx files
- Backend uses FastAPI, frontend is basic HTML/JS
- Agent creates necessary files including API documentation (Swagger UI) (2:24, 2:47)

### Cloud Deployment (Cloud Run & GKE)

- Agent automates deployment to Google Cloud Run first using provided scripts, then uses Terraform to deploy to a GKE cluster, handling cluster creation, secret configuration (OpenAI API Key), and deployment scripts (1:09, 6:22, 7:50)

### Kubernetes Optimization

- Agent analyzes current resource usage (CPU 94.1% limit) and implements priority fixes: adjusting HPA targets (CPU 70%->80%), increasing CPU limits (1500m -> 2000m), and optimizing readiness probes (60s -> 10s delay) (11:54, 12:24)

### Multi-Agent Configuration

- Users can enable 'Use Multiple Models' and select up to 4x agents (Composer 1, Sonnet 4.5, GPT-5.1 Fast, Grok 4 Fast) for parallel processing (13:00, 13:48)

### Workflow Control

- Agents operate within a 'Worktree' context separate from the main IDE, and users can switch between Agent mode and traditional Editor mode (1:51, 13:59)

![Screenshot at 0:00: The initial Cursor interface displaying four selected agents \(Composer 1, GPT-5.1 Fast, Gemini 2.5 Flash, Grok 4 Fast\) ready to tackle the 'Implement login feature' task simultaneously.](https://ss.rapidrecap.app/screens/m6cMChpRIgM/00-00-00.png)
![Screenshot at 0:05: Demonstration of the Bank Statement AI Analyser frontend interface, showing the file upload area and the 'Analyse Statement' button.](https://ss.rapidrecap.app/screens/m6cMChpRIgM/00-00-05.png)
![Screenshot at 0:11: A benchmark chart showing Cursor's 'Composer' model outperforming other models in both 'Intelligence \(Cursor-Bench score\)' and 'Speed \(Tokens per Second\)'.](https://ss.rapidrecap.app/screens/m6cMChpRIgM/00-00-11.png)
![Screenshot at 0:24: Slide introducing the new Cursor built for agents, emphasizing a shift from file-centric work to outcome-centric work.](https://ss.rapidrecap.app/screens/m6cMChpRIgM/00-00-24.png)
![Screenshot at 2:24: The agent planning the development of the AI insights application, listing 7 To-dos, including setting up project structure \(frontend, backend, database\) and creating the backend API \(FastAPI\).](https://ss.rapidrecap.app/screens/m6cMChpRIgM/00-02-24.png)
![Screenshot at 3:39: Agent configuring testing parameters in TestSprite, selecting 'Backend' mode and 'Codebase' scope for testing the application.](https://ss.rapidrecap.app/screens/m6cMChpRIgM/00-03-39.png)
![Screenshot at 4:07: Agent executing commands to get credentials and then running gcloud commands to build the Docker image, showing the deployment process starting.](https://ss.rapidrecap.app/screens/m6cMChpRIgM/00-04-07.png)
![Screenshot at 4:43: The detailed test report dashboard showing test case TC002 failing due to unsupported file formats, indicating the need for fixes.](https://ss.rapidrecap.app/screens/m6cMChpRIgM/00-04-43.png)
![Screenshot at 6:54: The deployment instructions detailing the three required items for securing deployment on GKE: GCP Project ID, OpenAI API Key, and gcloud CLI installation.](https://ss.rapidrecap.app/screens/m6cMChpRIgM/00-06-54.png)
![Screenshot at 11:13: The agent switching to the terminal to run git init, demonstrating the agent's ability to handle basic Git repository setup commands within the environment.](https://ss.rapidrecap.app/screens/m6cMChpRIgM/00-11-13.png)