Working with AI

During the track you used ChatGPT (and for some of you Claude or GitHub Copilot) to debug errors and explain code, and you wrote AI_ASSIST.md files about what you kept and what you changed.

A few rules to keep

Test AI-generated code the same way you test your own: run it, add a pytest test, check the output on a small sample.
Give the model context: the exact error message, the library version, a short sample of the data.
Ask for an explanation along with the fix, so you learn the concept and can solve it yourself next time. Better yet, let it create a concepts.md guide where you document these learnings (after reviewing it) for reference later on.
Be aware and mindful of which model you should choose for what work. Some models are very powerful but will blow through your daily/hourly limits quickly, and might be overkill for your task.
Remember that using LLM's is all about tokens/context. The more tokens you use, the more quickly you will blow pas your free-tier limits. Be mindful of the questions you ask and research best-practices/techniques to limit the amount of tokens per questions e.g. by using /compact in codex/claude.

AI at work

In this course we kept AI use simple: you asked an AI questions and reported on the answers. At work you may be asked to do much more with it. Many teams use coding agents that design and build a whole pipeline from a written description, skills and instruction files (such as AGENTS.md) that teach the agent the team's conventions, and AI reviews on every pull request. You stay responsible for the code the agent writes, so the rules above still apply.

The free plans below let you try most of these ideas on your own projects. Limits change often, so check each pricing page. Free plans may use your chats to train models, so keep work code and data out of them.

<aside> ❗ Do not use AI to do all your coding at this stage. You are still learning, and you learn best by practice. But be aware that your work might use AI differently than what you've done so far.

</aside>

ChatGPT and Codex

Codex is OpenAI's coding agent, which you likely used so far. Projects are free and let you keep instructions and a few files (for example your pipeline's README) together, so ChatGPT has the context of your project in every chat. Codex is a command line tool that you can run in any terminal, or you can use it inside the ChatGPT Desktop app. The Codex models are OpenAI's own, and are currently at the forefront of coding models.

Projects in ChatGPT: Free. How to set up a project with instructions and files.
Codex documentation: Free. How Codex works and how to give it tasks.
Custom instructions with AGENTS.md: Free. Write your project's rules down once, so the agent follows them every time.
Build skills: Free. Package a repeatable workflow, such as "add a dbt model with tests", as a skill.
OpenAI Academy: Free. OpenAI's own courses and live sessions on using ChatGPT well.
OpenAI prompt engineering guide: Free. Official tips with examples.
Codex pricing: what each ChatGPT plan includes.

GitHub Copilot

GitHub Copilot is a bit different from ChatGPT and Claude since it's owner (Microsoft) doesn't provide their own models but routes requests to other Provider's models such as ChatGPT (OpenAI), Claude (Anthropic) and others. That's why you can't really judge Copilot solely on its answers, it really depends on the Provider's model you choose, since Copilot simply is are wrapper for them.

Copilot Free works inside VS Code, which you used during the track. It includes agent mode and custom instructions, with a monthly limit on chat and agent requests. Pull request reviews need a paid plan.

Use chat in VS Code: Free. Ask, edit, and agent modes in VS Code.
Custom instructions in VS Code: Free. Project rules for Copilot, similar to AGENTS.md.
Agent Skills in VS Code: Free. The same skills format that Codex uses.
About Copilot code review: Free to read. How teams let Copilot review pull requests.
GitHub Copilot Fundamentals: Free. Microsoft Learn path on Copilot in VS Code.
GitHub Copilot best practices: Free. What Copilot is good at and where to be careful.
Copilot plans: what Copilot Free includes.

Claude

Claude Code is Anthropic's coding agent cli tool. Claude Code is very powerful and widely regarded as perhaps currently the most advanced and feature-rich coding agent framework, competing closely with ChatGPT Codex. However, there is no free trial and it requires a paid plan that starts at around 20e per month. If you are not willing to pay for Claude Code, then stick to Codex. Or perhaps try a single month to see if it's worth the cost for you.

Use skills in Claude: Free. How to turn on and use skills.
Claude Code overview: Free to read. What a coding agent in the terminal looks like.
Anthropic Courses: Free. Courses on using Claude, including prompting and coding with AI.
Anthropic courses on GitHub: Free. Open-source notebooks on prompt engineering. The ideas apply to ChatGPT too.
Claude pricing: what the free plan includes.

General reading

These apply to any AI tool.

Here's how I use LLMs to help me write code by Simon Willison: Free. A practical, honest write-up by an experienced developer.
Building effective agents: Free. Explains what an AI agent is and the common patterns behind agent tools.
Agent Skills: Free. The open format behind skills in Codex, Copilot, and Claude. A skill you write for one tool mostly works in the others.

The HackYourFuture curriculum is licensed under CC BY-NC-SA 4.0 https://hackyourfuture.net/

CC BY-NC-SA 4.0 Icons

Built with ❤️ by the HackYourFuture community · Thank you, contributors

Found a mistake or have a suggestion? Let us know in the feedback form.