Clase 01 - Asistentes de IA desde cero con Claude Cowork y Gemini Gems

Summary

The class introduces the progression from generative AI to AI assistants and AI agents:

• Generative AI creates content in response to prompts.
• AI assistants combine a generative model with specific knowledge and instructions, but still rely on a person to ask questions.
• AI agents are more autonomous: they can use tools, make decisions, and carry out multistep tasks toward a goal.

The instructor emphasizes that an AI tool’s usefulness depends heavily on how clearly it is instructed. The class’s prompting framework is the IDEAL method:

• I — Intention: State the goal.
• D — Details: Provide relevant context and information.
• E — Examples: Show the desired answer style or format.
• A — Action: Specify the task the AI should perform.
• L — Limits: Set restrictions, tone, length, or other boundaries.

A job-interview example contrasts a vague request with a detailed prompt. The more specific version produces a tailored preparation plan, priority topics, and sample questions. A contract example demonstrates using AI to summarize clauses, flag potential issues, and draft a polite email requesting changes. The instructor cautions against uploading sensitive personal information and stresses that AI output should be reviewed, especially for legal matters.

Product Demonstrations and Practical Guidance

Gemini Gems

The class demonstrates creating a custom assistant in Gemini, called a Gem. Setup includes a name, description, and instructions built with the IDEAL framework. Users can add organizational reference documents as a knowledge base, test the Gem’s responses, revise its instructions or materials, save it, and share it through a link.

The example is a coffee-shop staff-training assistant that answers questions using internal procedures, service standards, menu information, and allergy guidance. It is intended to help standardize answers and processes across staff or locations. The instructor recommends documenting processes and testing assistants because AI can make mistakes or respond incorrectly.

Claude Cowork

The desktop app is shown as a way to delegate work involving files on a computer. The user selects a specific folder and grants access, rather than giving the tool unrestricted access.

Demonstrations include:

• Sorting downloaded documents into subfolders by file type.
• Reviewing invoices in a selected folder and creating an Excel expense summary with dates, amounts, suppliers, and descriptions. The demonstration notes that totals in different currencies are kept separate rather than combined.

The instructor recommends using manual mode when first learning the tool so its steps can be reviewed, then considering automatic mode once a workflow is understood. Users can also select models according to task complexity and monitor the tool’s actions and permissions.

Main Takeaways

• Apply AI to specific workflows rather than relying on vague, generic prompts.
• Build assistants around documented, relevant knowledge and test them before sharing or relying on them.
• Review AI-generated work. The class describes AI as completing much of a task, while human judgment remains necessary for verification and decisions.
• Learn the underlying concepts rather than becoming dependent on one product, since interfaces and features change.
• A later lesson will preview building an automated agent with n8n, including an example of an AI agent replying to an email. That workflow is not taught in detail here.

The presentation also cites market-demand statistics and promotes a broader training program; those claims are part of the speakers’ introduction rather than demonstrations in the lesson.

Main Speakers and Sources

• Ingrid Estupiñán — Daxus instructor and primary presenter; leads the conceptual explanations and tool demonstrations.
• Saira Hurtado — Daxus co-presenter, identified in the class as an Excel and Power BI instructor and Daxus Latam CEO; participates in hosting and audience interaction.
• Tools demonstrated or discussed: Claude, including Claude Cowork; Google Gemini Gems; and n8n as a preview for the next class.