Video summary

How To Learn To Code In 2026

Main summary

Key takeaways

Technology

Summary: How to Learn to Code in 2026 (Tech-Focused)

Learning to code in 2026 is portrayed as different from earlier eras, largely due to AI coding tools and a shift toward LLM-driven workflows. Even with AI acceleration, the video emphasizes that you should still learn coding fundamentals—because AI agents require human oversight to guide work, validate correctness, and catch blind spots, especially around security and privacy.

AI can move fast, but it still needs humans to ensure correctness and safety.

Key Concept: “Vibe Coding” vs. “Agentic Engineering”

The summary cites Andrej Karpathy’s terminology:

  • Vibe coding Giving in to “vibes” with LLMs, where you often don’t deeply reason about the code. This is framed as more suitable for throwaway projects.

  • Agentic engineering / programming via LLM agents A more professional, orchestrated workflow where multiple AI coding agents collaborate under human oversight.

Important Warning: Orchestrating Agents Requires Fundamentals

To successfully orchestrate multiple agents, you need engineering fundamentals:

  • Understand the system/design level of what you’re building.
  • Provide context and documentation that agents can use.
  • Know what’s correct so you can detect and correct agent mistakes.

Because of this, the learning focus should shift from line-by-line mechanics toward big-picture software design, systems, and architecture, with strong emphasis on security and privacy.

Recommended Learning Plan (Roughly Chronological)

1. Basics of Coding

Key topics include:

  • Variables, types
  • If statements and loops
  • Object-oriented programming (OOP)
  • APIs

Language guidance mentioned:

  • Python for building AI agents and data-related work
  • JavaScript for web app development

Emphasis: fundamentals are necessary to read and understand AI-generated code. The summary also recommends prioritizing human-made/vetted resources over AI-only code, noting that AI-generated code ultimately traces back to human-written code.

2. Software Architecture / Systems

Topics include:

  • Project structure
  • Choosing a tech stack
  • System design
  • API choice
  • Data flow
  • Storage choices, including databases
  • Testing
  • Deployment

Rationale: to direct agents effectively (“build this, redirect when wrong”) and to conceptualize the final product from requirements—for example, choosing the right database for a high-query system.

3. Version Control & GitHub

Topics include:

  • Git for tracking changes
  • GitHub for sharing/collaboration

Why it matters for agentic engineering:

  • Track what agents changed
  • Revert mistakes
  • Avoid losing the codebase

4. Security and Privacy (Separate, High Emphasis)

The summary flags security/privacy because AI coding agents may have blind spots.

Guidance: don’t assume agents will handle it. Be explicit about principles like authentication/integration, even though it’s sometimes treated as “embedded” in other topics.

5. Bonus: Microservices / Containerization (Optional but Recommended)

This section describes packaging apps with dependencies in containers to isolate them from the host environment.

Benefit with AI agents:

  • If an agent “runs amok,” the damage can be contained.
  • You can shut down the isolated environment.

It’s described as more advanced, but useful for safety and reliability in agent-driven development.

6. Then: Agentic Engineering Itself

Agentic engineering is treated as a skill involving:

  • Setting up projects and AI coding agents
  • Communicating requirements to agents
  • Monitoring multiple agents
  • Orchestrating different agents together

Resources may be limited because it’s relatively new (per the summary).

Tools / Products Mentioned (Review-Style)

Warp + Oz (Sponsored)

Warp

  • Described as a pro developer tool across the full AI coding spectrum.
  • The user claims they built a project with it.

Oz

  • Described as a cloud-based agent orchestration platform built on top of Warp.
  • Multi-agent example: one agent for backend, one for documentation, one for tests.
  • Oz dashboard: real-time visibility into agents’ status/issues.
  • Isolation: each agent runs in its own isolated environment to prevent interference (“keeps the MacBook Air alive”).
  • Example feature: a scheduled agent that updates a microsite daily using AI-generated news summaries.
  • Integrations/workflow tools: convert skills into agents, integrate with Slack and Linear.
  • Claimed stats: 700,000 engineers using it and 97% of code diffs accepted.

Learning Speedups: Tips to “Supercharge” Learning

Using AI to Improve Study

  • NotebookLM Ingest course/text/video material, then ask for summaries and questions to guide learning.

  • Claude for code explanation (recommended default) Alternatives mentioned: Gemini / other modern models and free/open options like “Gwen, Kimmy, DeepSeek” (names may be imperfect due to subtitle errors).

  • Ask for examples and analogies to solidify concepts (e.g., explaining OOP with analogy).

  • Use AI to explain existing code snippets:

    • upload/copy code
    • ask for structure and line-by-line meaning
    • optionally set up an AI coding agent to explain

Projects as the Main Learning Engine

The summary strongly claims projects are “hands down” the best way to learn.

Guidance:

  • Build projects aligned to what you’re studying:
    • APIs → use APIs
    • testing → write tests
    • learn stack differences by trying them
  • Example agent-assisted project prompt: a music storage project using:
    • Django backend
    • Next.js frontend
    • SQLite
  • Other project learning methods:
    • adopt and modify others’ projects (add features, change code, observe outcomes)
    • ask AI for multiple implementation approaches
    • use AI to interpret documentation faster (speaker dislikes reading docs directly)

Vibe Coding Tips

Not covered deeply; it only references that “vibe coding” tips may be separate and potentially confusing without experience.

Main Speakers / Sources

  • Speaker: Tina

    • Former data scientist at Meta
    • “Coding for 10 years”

  • Source cited: Andrej Karpathy Referenced via posts defining “vibe coding” and discussing LLM agent default workflows.

  • Sponsored tools/products: Warp and Oz Oz is described as an orchestration platform built on top of Warp.

Original video