Skip to content

Instantly share code, notes, and snippets.

Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save cedrickchee/984d8839df32c2469f4da3877878ebfc to your computer and use it in GitHub Desktop.
Save cedrickchee/984d8839df32c2469f4da3877878ebfc to your computer and use it in GitHub Desktop.
What makes a true AI coding assistant?

What Makes a True AI Coding Assistant?

What is an AI Coding Assistant?

If the coding assistant can't run ITERATIVE CRUD on ALL of your code, it's not a True AI Coding Assistant (TACA)

Standards for True AI Coding Assistants

  1. Must work on existing codebases
  2. Must have a file context mechanism
  3. Must be iteratively controllable

Examples

  • CrewAI (agentic building block - not TACA)
  • Autogen (agentic building block - not TACA)
  • GPT Engineer (rapid prototyping - not TACA)
  • ChatDEV (rapid prototyping - not TACA)
  • Cursor (Legitimate TACA)
  • AIDER (Legitimate TACA)

Best two AI Coding Assistants as of 2024 January

  • Aider

  • Cursor

    Video: Not a cherry-picked Copilot++ demo

    From level 0 to 5: the steps to automate software engineering - how engineers are using AI tools, how they can progress, and how such tools can revolutionize the way we code.

    • Level 4: Enabling you to think less about individual lines of code. They enable you to think about the bigger picture, and give you the ability to up level your perspective.

    • Level 5: Unclear at the moment. The AI tool behaves like an agent? And incorporate several tools? Devin? Devika? Sweep? LTM-1?

5 BIG problems with AI Coding Assistants

  1. File Management
  2. Accuracy
  3. Speed
  4. Security
  5. Skill Atrophy 💰💰💰💰💰

Potential Solutions to the 5 BIG problems

  • File Management
    • Explicit manual add/remove file context
    • Wait for tools to improve
    • Look for niche solutions that include open file context
  • Accuracy
    • PRACTICE
    • The prompt is the new fundamental unit of programming
    • Wait for tools to improve
  • Speed
    • Experiment and see if you REALLY need gpt-4-turbo or if gpt-3 is enough
    • Wait for tools to improve
  • Security
    • Bite the bullet and use a cloud solution
    • Use a local-only solution (in the future)
      • Waiting for great local LLMs to catch up
  • Skill Atrophy
    • Great technology always destroys skill, jobs, and entire industries
    • Pay close attention to the results of your prompts to slow skill decay
    • Don't focus on this too much, focus on the bigger picture: The Prompt
    • Flow with the AI, don't fight it

Why care about AI Coding Assistants?

  • Whoever cracks this puzzle first is going to catapult into the future of engineering.
  • The world is built on code, and whoever can build the best code the fastest will win in the long run.
  • If you're faster, you can build more, once you finish your work you can work more or relax more.
  • There is a massive first movers advantage in the software industry. Learning and mastering AI Coding Assistants ASAP gives you a monumental edge.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment