Thinking partner

Use GPT for understanding, planning, and practice before the agent edits.

ChatGPT and GPT models are excellent for tutoring, brainstorming, explaining errors, writing specs, and exploring options. They are not a substitute for a repo-aware coding agent unless you give them the relevant files and verify the output locally.

When you'll want this You need a patient tutor, a second opinion, a spec draft, or a way to rehearse a task before asking an agent to change files.

#Where GPT fits in Groundwork

Learn a concept	Ask for a beginner explanation, examples, quizzes, and a tiny exercise. Then apply it in the terminal or editor.
Write a spec	Turn a fuzzy idea into requirements, constraints, acceptance checks, and a first implementation plan.
Compare approaches	Ask for tradeoffs before you let a coding agent touch the repo.
Debug your thinking	Paste an error, command output, or a small code snippet and ask what assumptions could be wrong.
Create learning drills	Have GPT generate practice loops for shell, Git, Unity, React, Python, or any track you are learning.

Boundary

Do not treat a chat answer as verified. If it affects a real repo, turn it into a local task for Codex, Claude Code, or another agent, then review the diff and run the checks.

#Prompts that teach instead of just answer

tutorTeach me this concept like I am new to it, then give me three tiny exercises and the answer key separately.

specTurn this idea into a one-page implementation spec with goals, non-goals, files likely involved, tests, and edge cases.

review before agentHere is my plan. Find unclear requirements, risky assumptions, and what I should ask the coding agent to verify before it edits.

after agentHere is the diff summary and test output. Explain what changed, what risk remains, and what I should manually inspect.

#GPT vs a coding agent

GPT / ChatGPT

Best for thinking

Clarify the idea, learn the basics, compare options, draft prompts, and create practice exercises.

Codex / Claude Code

Best for changing repos

Read project instructions, inspect files, edit, run commands, show a diff, and verify the result in the real workspace.

#Daily practice

Ask for a concept

Pick one term from the glossary and ask GPT for a plain-English explanation plus a command you can run.

Draft an agent prompt

Ask GPT to improve a vague coding request into a precise agent task with acceptance criteria.

Verify locally

Give the improved prompt to Codex or Claude Code in a small repo. Read the diff and compare it to the spec.

#Official references

ChatGPT — OpenAI's main chat product.
OpenAI ChatGPT help — official help articles for ChatGPT features and account behavior.
OpenAI platform docs — developer docs for using GPT models through the API.
Codex UI & CLI — when the answer needs to become a repo change.

← PrevCodex UI & CLI Next →GLM & open models