Video summary

How I'd learn to code if I had to start over

Main summary

Key takeaways

Educational

Main ideas and concepts (lesson summary)

  • The video presents a step-by-step “roadmap” for learning to code from scratch, emphasizing that coding can feel scary at first—but becomes easier with practice and structure.
  • It uses a cooking analogy: learning to code is like learning to cook. It’s confusing at first, but becomes clearer when you have guidance and start doing.
  • The core methodology is 70-25-5 time allocation:
    • 70% build
    • 25% consume
    • 5% AI experimentation
  • Over time, the approach calls for progressively more real projects and stronger tooling/skills.

The overall goal: move from passive learning to active building with consistent feedback loops.

Methodology / instructions (detailed)

Step 1: Pick a programming language (foundation choice)

  • Choose a beginner-friendly language that won’t overwhelm you.
  • Recommended (if starting over): Python
    • Why: simple, clean, readable, widely used
    • Also relevant for AI/ML because many AI libraries use Python
  • Optional early path: Scratch (Harvard CS50)
    • Purpose: builds basic computing foundations
    • Guidance: don’t spend more than 2–3 weeks on Scratch
    • Transition: move to Python afterward for meaningful progress

Notes on other languages (alternatives)

  • JavaScript: can be easier for some, but may confuse others due to “weird behaviors.”
  • Java: likened to slow cooking—verbose and patience-requiring, but builds discipline.
  • C: very challenging like making a soufflé from scratch; hard, but builds strong expertise if you persist.

Step 2: Find the right resources + avoid the “illusion of competence”

Key pitfall: “illusion of competence”

  • Watching tutorials can make you feel capable.
  • The moment you try on your own, you realize you don’t know how to solve problems.

Fix: apply the 70-25-5 rule

  • 70% build
    • Build coding projects (this breaks the illusion quickly because it reveals your gaps)
    • Benefits:
      • concrete examples for abstract concepts
      • real accomplishment
  • 25% consume
    • Learn new information via resources
    • The goal: upgrade what you’re building (not just collecting knowledge)
  • 5% experiment with AI tools
    • Do this sparingly because fundamentals come first
    • Beginners probably should not rely on AI heavily
    • Over time, this share could increase (potentially 20–30% depending on comfort)

Tool/resource recommendation

  • cody.tech
    • Bite-size lessons with: lesson + quiz + project per module
    • Includes an AI assistant to help when you’re stuck
    • Mentions free access and a discount code (as shown in the video)

Step 3: Build end-to-end, real-world projects

  • Move beyond small exercises into legit full projects that connect:
    • front end
    • back end
    • integration/glue code
  • Emphasis on relevance/opportunity:
    • AI security is growing because companies adopt AI quickly while security training lags

Example project idea: full-stack weather dashboard

  • Front end: user enters a city; UI displays results
  • Back end: Python with Flask to fetch weather data
  • Database: MongoDB to store favorite cities
  • Front end framework: React to present data cleanly
  • Learning outcome: repeated reps with full-stack integration (front end + back end + glue code)

Additional project ideation support

  • Use a GitHub repository called “app ideas”
    • Contains many real project ideas grouped by beginner/intermediate/advanced tiers
    • Helps you keep moving to “new dishes” as you level up

Step 4: Master coding tools (be “top 1%”-style proficient)

  • Uses a “chef” analogy:
    • Pros aren’t just better because of secret recipes—they have superior tool mastery (“knife skills”).

Advice from an ex-Google software engineer (quoted)

They focused on:

  1. API architecture
  2. life cycle and version control
  3. a proper tech stack

Acronym: ALT (three skills)

  • A = API architecture
    • An API is structured communication between systems.
    • Analogy: a waiter takes your order and brings back the dish.
    • In software, the front end doesn’t directly access the database—it requests via an API.
    • Examples mentioned:
      • Google Maps (location data)
      • Amazon (customer data)
  • L = life cycle + version control
    • Specifically: Git and GitHub
    • Important for collaboration with other engineers
  • T = tech stack
    • Database (e.g., MongoDB) = pantry (ingredients storage)
    • Back end frameworks (e.g., Flask or Node) = stove (core logic execution)
    • Front end frameworks (e.g., React, Angular, Vue) = plating (presentation)

Step 5: Learn coding using AI (but only after learning fundamentals)

Strong warning

  • Don’t use AI as a beginner to generate entire code files for you.
  • Instruction: learn coding properly first, then use AI to augment learning.

Three levels of approaching AI tools

  • Level 1: Autocomplete (tool: GitHub Copilot)
    • For small tasks (e.g., completing function logic, finding specific utilities)
    • Goal: speed up work you already understand, not replace learning
  • Level 2: Generation (tools: Cursor or Windsurf)
    • Prompt in plain English to generate entire functions/files
    • AI can edit across your code base
    • You remain responsible for steering, reviewing, and verifying correctness
    • Analogy: AI acts like a sous chef for bigger chunks
  • Level 3: Delegation (tool: Claude Code)
    • Hand off entire features/projects
    • AI runs tasks autonomously: writes across multiple files and runs commands
    • Mentions “agents” splitting work (e.g., front-end agent, back-end agent, tests agent)

Final caution

  • Keep fundamentals and judgment.
  • Real-world shipping requires expertise and strategic decision-making even with AI assistance.

Featured speakers / sources (identified at the end)

  • Sacha (main presenter; “My name is Sacha”)
  • An ex-Google software engineer (quoted advice: API architecture, version control/life cycle, tech stack)
  • TryHackMe (referenced as a hands-on AI security/cybersecurity training platform; includes an example “AI security learning path”)
  • cody.tech (Python learning platform with mini lessons/quizzes/projects + AI assistant)
  • GitHub Copilot (recommended for AI autocomplete)
  • Cursor and Windsurf (recommended for AI code generation)
  • Claude Code (recommended for AI delegation)
  • Harvard CS50 (referenced for Scratch learning option)
  • GitHub repository “app ideas” (collection of project ideas)
  • Gordon Ramsay (used as an analogy source in the “illusion of competence” example)

Original video