Guía de Cowork: el asistente de IA que trabaja en tu ordenador (mejor alternativa a OpenClaw 🦀)
Introduction to Cowork AI Assistant
Overview of Cowork
- Cowork is an AI assistant developed by Cloud, designed to work on your computer and browser, capable of handling complex tasks by breaking them down into simpler ones.
- The video aims to demonstrate how easy it is to use Cowork and how it can simplify daily tasks.
Comparison with OpenCl
- OpenCl has popularized the demand for AI assistants that can manage various tasks like booking flights and organizing emails. However, there are security concerns regarding sharing sensitive information with experimental AI agents.
- The developer of OpenCl has partnered with a leading security firm to address these vulnerabilities, but users are advised against using it unless they are highly technical.
Installation and Features of Cowork
Installation Process
- Installing Cowork is straightforward; users can download the application for both Mac and Windows from the Cloud website. After installation, registration with a paid Cloud account is required.
User Interface
- The interface includes three main buttons: Chat (similar to Cloud's web app), Code (for programmers), and Cowork (the focus of this demonstration). Users can access different functionalities through these options.
Practical Examples Using Cowork
Organizing Desktop Files
- A practical example involves asking Cowork to organize a cluttered desktop by selecting a prompt and model (recommended: Opus 4.6) for optimal performance. Users must grant permission for file modifications on their desktop before proceeding.
- Once initiated, Cowork analyzes existing files on the desktop, asking how users want them organized—by project or file type—and whether to rename confusing files based on their content analysis. This showcases its ability to handle complex tasks efficiently by breaking them down into manageable steps.
Task Execution
- As it processes the task, Cowork creates folder structures and renames files accordingly, demonstrating its utility in quickly organizing digital spaces without user intervention. This feature significantly enhances productivity by automating mundane organizational tasks within minutes.
Searching for Flights with Cowork
Flight Search Example
- Another useful application of Cowork is utilizing it to search for flights online; in this case, finding connections from Granada to San Sebastián while avoiding Madrid as a layover point.
- Users specify preferences such as analyzing multiple airports without focusing solely on price since the trip will be covered by the company budget.
This structured approach provides clarity about what features and capabilities make the Cowork AI assistant valuable in everyday scenarios while highlighting potential risks associated with similar technologies like OpenCl.
Analyzing Flight Options with AI
Task Breakdown
- The task involves analyzing nearby airports, searching for flights, comparing options, and summarizing the best choices. Cloud is utilized to navigate Google for this analysis.
Permissions and Control
- Cloud takes control of the browser, requiring continuous permission to perform actions like writing or navigating. This raises concerns about security and user consent.
Risks of AI Interaction
- A significant risk associated with AI systems connected to the internet is prompt injection attacks, where malicious inputs can manipulate AI behavior.
- Prompt injections can lead to unauthorized actions such as revealing confidential data or executing harmful code by embedding hidden instructions in web pages.
Limitations of Current Models
- Three main limitations of current AI models are discussed:
- Availability: Currently only accessible on Mac systems.
- Speed: Real-time analysis may be slower than manual tasks but allows users to multitask.
- Usage Limits: Users may encounter limits on model usage due to token consumption.
Recommendations for Efficient Use
- Users should monitor their usage limits in settings and consider using less powerful models for simpler tasks to conserve tokens.
- In paid plans like Pro, usage limits reset every four hours, allowing continued access after reaching a limit.
Exploring Flight Options
Flight Analysis Results
- After analyzing various flight options from Málaga to Bilbao, several routes are identified including direct flights and those with layovers in cities like Sevilla and Barcelona.
Researching AI Model Benchmarks
Benchmark Analysis Request
- A request is made for Cloud to identify key benchmarks used for evaluating major AI models and gather their performance metrics.
Generated Excel Report
- Cloud generates an organized Excel file containing benchmark details categorized by model families (OpenAI, Google, Anthropic), providing insights into each model's performance metrics.
Summary of Findings
- The report includes descriptions of benchmarks along with classifications for various models. It highlights that some benchmarks may not be publicly available but provides sufficient information on known models.
Benchmarking AI Models and Presentation Creation
Overview of Model Rankings
- The discussion begins with an analysis of how different AI models rank across various benchmarks, highlighting that GPT 5.2 is currently the leader in one specific benchmark.
- A request was made for the generation of an Excel file to organize this ranking information visually, emphasizing the simplicity and efficiency of such a task.
Creating Impactful Presentations
- The speaker describes a scenario where they need to prepare a brief yet visually impactful presentation within ten minutes, focusing on key benchmarks and model rankings.
- Upon executing the request, a dynamic presentation is generated that includes essential elements like benchmark comparisons, results by provider, and final rankings.
Key Insights from Generated Presentation
- The presentation outlines significant results for each provider: GPT 5.2 leads with impressive scores followed by Gemini 3 Pro and others like Antropic and Tipsic.
- It emphasizes the ease of editing within the presentation format, allowing for quick corrections or updates as needed.
Efficiency in Information Management
- The speaker reflects on how quickly they were able to gather information and create a presentation, showcasing the potential for delegating tasks related to data collection and organization efficiently.
- They express gratitude towards viewers for their support while encouraging questions or clarifications in comments.