Video Summary - [Aula 2] Como transformar o Claude no seu clone

Main ideas / lessons (Lesson 2: turning Claude into “your clone”)

The video explains how to transform an AI assistant (Claude / “Cloude”) from a generic chatbot into a “clone” that can work like you or your company. It focuses on training the AI using:

Projects (what the AI “knows”)
Skills (how the AI “does” tasks, i.e., execution style)
Connectors / data access (where the AI “looks” for information)

A major theme is to stop repeatedly explaining the same context and instead make the AI follow your company’s rules, writing style, and workflows automatically.

Methodology / step-by-step instructions

A) Three “pillars” to remove generic behavior

What the AI knows (company context) Teach it about the company: what it does, who clients are, and how operations work. Implementation mechanism: Project instructions + saved context.
How the AI works (execution style) Teach the company’s way of doing tasks—writing emails, assembling presentations, and making decisions. Implementation mechanism: Skills (stored reusable “how-to” documents).
Where the AI searches (information sources) Provide access to documents and systems it needs (Drive, email, spreadsheets, etc.). Implementation mechanism: Connectors / file access.

B) Create a Project to teach “what the AI knows”

Goal: Teach the company basics once, store them in a project, and avoid re-explaining later.

Steps described:

In Claude’s interface, open Projects
Click + (create new project)
Name the project (example: “Praxis”)
Inside the project, fill two main parts:

1) Instructions

A short “general company description” and guidelines (role, services, vocabulary preferences, etc.)
The video warns that instructions should be “basics,” not an enormous essay.

2) Files / Context (sometimes called “context”)

Upload or provide documents containing detailed company information

(Important advanced note) Enable “memory” for the project

Go to Profile → Settings → Memory
Turn on options so the AI can remember useful past conversation details inside that project.

C) Validate the “Project knowledge” test

Test prompt example (paraphrased from the video):

Ask the AI to write a short company introduction paragraph for the footer of a business proposal.

Expected behavior:

Before training: the AI requests missing info.
After training: the AI produces the paragraph using the saved project context.

D) Teach “how the company works” using Skills

Goal: Turn your company’s execution rules into reusable “abilities” the AI can apply consistently.

What a Skill is (as explained):

A stored document (“notepad”) that encodes how to do a task.
Skills are saved to Claude and can be reused across projects/conversations.

Types of skills mentioned:

Email writing style
Presentation assembly in your brand format
Month-end report closing
Decision-making procedure (executing policy rules)

How to build a Skill (two approaches):

Directly instruct the Skill to be created Provide the decision-making rules / workflow and explicitly tell the AI to turn it into a skill.
Refine through normal conversation Chat normally so the AI drafts/iterates. When the output is excellent, ask it to convert the accumulated learning into a reusable skill.

Key rule emphasized: When requesting a skill, you must explicitly say something like “turn this into a skill”; otherwise it may not store it correctly.

E) Create a decision-making Skill (core example)

The video demonstrates training Claude to make decisions using a company policy with five rules, such as:

Charge fixed price per project (not percentage of spend)
Don’t accept work without real data access
Don’t make promises
Follow a no exclusivity / exclusivity rule (example shows exclusivity being rejected)
Don’t accept offers that conflict with company constraints

Process shown:

Provide:
- Rules + decision criteria + examples (“precedents”)
Ask Claude (often inside the project context) to:
- Create a skill for those decisions
Download / save the resulting Skill file
Use it later by invoking the skill (example method mentioned: using “/skillname”)

F) Test Skills in real scenarios

After training, the “proposal decision” scenario is tested twice:

Without skill The AI may recommend options that violate policy.
With skill The AI applies rules immediately (e.g., refuses exclusivity offer, rejects percentage pricing) and outputs a structured recommendation.

G) Build a business proposal “our way” (presentation Skill + iteration)

The video then shows creating a complex client proposal slide deck (HTML-based), including:

Company problem summary
Project phases (e.g., diagnosis → implementation → assisted operation, etc.)
Pricing, validity, deliverables
Simulations and interactive visuals
Styling requirements (fonts/colors/graphics)

Workflow shown:

Prompt Claude to assemble the proposal with the given company metrics and requirements
Iterate:
- If something is visually or structurally wrong, correct and regenerate
Convert the conversation into a Skill:
- Next proposals can be generated quickly using the saved template + learned execution style

H) Make presentations interactive / dynamic (drag bar, colors, transitions)

Demonstrations include:

A draggable simulation bar for conversion rate → payback timeline
Color-coded regions:
- Green when the project pays back
- Red when it doesn’t
- Yellow/amber for borderline outcomes
Slide animation / gradual building effects
Improved readability via:
- typography, rounded cards, transparency/glass effects, spacing, and alignment

The presenter emphasizes:

Keep improving until it feels “perfect,” then turn the best conversation into a skill.

I) Use Connectors for “where Claude looks”

Goal: Instead of uploading files repeatedly, let Claude access your tools.