Lesson 3 of 4

AI-Assisted Development Tooling

Onboarding onto a codebase with an AI pair-programmer for the first time changes what "reading code" and "writing code" actually mean day to day.

The tools, by autonomy level

Four tools cover most of the space, from least to most autonomous.

GitHub Copilot

Inline, autocomplete-style suggestions inside your existing editor — the lowest-friction entry point, least autonomous.

Cursor

An AI-native editor built around multi-file edits and chat-driven changes inside a familiar IDE.

Claude Code / Codex (CLI agents)

Terminal-based agents that can read a whole repo, run commands, and make multi-file changes autonomously, with the developer reviewing rather than typing every keystroke.

ChatGPT (general assistant)

Used outside the editor entirely — for explaining an unfamiliar error, drafting a script, or reasoning about an approach before writing any code.

What actually changes about your workflow

Three shifts matter more than which specific tool you pick.

Review discipline

The bottleneck shifts from writing code to reviewing it — read every diff as critically as a teammate's PR, not a rubber stamp because "the AI wrote it."

Prompting as a skill

Vague prompts get vague code. Specifying constraints, file paths, and the pattern to follow up front saves more time than iterating on a bad first attempt.

Trust boundaries

Treat AI-generated code touching auth, payments, or data deletion with more scrutiny than a boilerplate change — the tool doesn't know which lines are load-bearing.

Exercises

Compare the two prompts below, then work through both exercises.

Vague: "fix the bug in the login form"

Specific: "In LoginForm.tsx, the submit handler doesn't disable
the button while the request is in flight, so a slow network
lets a user double-submit. Add a `pending` state and disable
the button while `pending` is true — follow the pattern already
used in SignupForm.tsx."

Concept check

Answer in your head first, then click a question to reveal the answer.

Scenario checklist

Work through the sample prompt comparison and check each item off. Progress is saved in your browser.

Identify what's actually vague or underspecified about the first promptAdd the missing constraints: file, symptom, and existing pattern to followCompare what each version of the prompt would plausibly produceIdentify what you would still need to review in the specific version's outputDescribe when you would escalate from autocomplete to a full agentic tool for this fix

Go deeper

These primers are a starting point. Read the primary sources to build real fluency.

GitHub Docs: Quickstart for GitHub CopilotThe lowest-friction entry point into AI-assisted coding, from the source Cursor Docs: QuickstartHow multi-file, chat-driven editing actually works in practice Claude Docs: Claude Code overviewA terminal-based agent that reads a repo, edits files, and runs commands

Training Notes

What building this project taught me.

This module isn't a flagged gap, but it changes how the others get done

Full-Stack Production Bar and every other module in this curriculum will likely get built with one of these tools — this lesson is about using them well, not just fast.

The vague/specific example is a real pattern, not an extreme case

Most low-quality AI-generated code traces back to an underspecified prompt, not a model limitation — the fix is almost always adding the missing constraint, not switching tools.