AI Coding Tools

Plan first, tool second

Do not choose a coding AI before you know what you are building. A tool chosen before you have a Sprint Goal and Kanban tasks is a tool chosen for the wrong reasons.

Come back to this page after Mission 4 in the Week 4 session guide.

---

Tools available to you

All tools below are free (or have a meaningful free tier) and have been assessed for availability from mainland China.

Tool	Cost	Works in China?	Best for	Watch out
GitHub Copilot	Free (student)	✓ via VS Code	Inline completions, scaffolding new files, multi-file agent mode	Supervise everything it writes — you explain it at Checkpoint
Qwen Code	Free (1,000 req/day)	✓ Alibaba-hosted	Well-scoped, specific tasks with clear inputs and outputs	Weaker on open-ended or ambiguous tasks — define the task precisely before using it
TRAE	Free	✓ ByteDance	UI prototyping, image-to-code, Builder mode; SOLO mode for non-coders (describe visually, AI builds)	Made by ByteDance — data practices differ from other tools. Fine for coursework; avoid for anything sensitive or proprietary.
Windsurf	Free tier (generous)	✓ mostly⁴	Full-project agent that reasons across files; good Cursor alternative on campus	Free tier has monthly limits — track usage
Cursor 3.0	Free tier only³	✗ mainland China²	Parallel agents, Design Mode (point at UI), /best-of-n	Does not work from campus — U.S. models restricted in mainland China; free tier cannot use China-accessible models
Claude.ai / ChatGPT / DeepSeek	Free tier	Varies¹	Explaining code, debugging, planning, writing prompts	Not agentic — you paste in, you copy out. Slower but more controllable.

¹ Standard browser required. WeChat in-app browser does not reliably support these tools.
² Cursor restricts U.S.-hosted models (Claude, GPT-4, Gemini) in the region. Off-campus with international network access: it works and is worth using.
³ Student Pro (one free year) requires a university email — XJTLU (.edu.cn) addresses are not accepted. The free tier cannot use China-accessible models (Qwen, DeepSeek) — bringing your own API key is Pro-only.
⁴ Windsurf uses Codeium’s own servers and is accessible from most Chinese networks, but availability can vary by ISP. Test on your connection before relying on it.

GitHub Copilot has a dedicated setup guide — see GitHub Copilot (free) if you have not claimed your student access yet.

Cursor 3.0 has a dedicated page — see Cursor 3.0 for features, pricing, and the China limitation in detail.

---

The evaluation exercise

Don’t pick a tool by reading about it. Pick a tool by trying it on a real task.

Pick a task first

Take one specific, bounded task from your Kanban board — something with a clear input and expected output. Run it with two tools. Compare.

	Tool A	Tool B
Did it understand the task?	Yes / Partly / No	Yes / Partly / No
Did the code work first try?	Yes / After fixes / No	Yes / After fixes / No
Can you explain what it wrote?	Yes / With help / No	Yes / With help / No
Time to working result

Record your findings in your Dev Log under this week’s entry. One paragraph is enough.

What “partly” tells you

If a tool only partly understood the task, the gap is usually in how the task was described — not the tool. Try rewriting the task description before switching tools. Being precise about inputs, outputs, and constraints is a skill in itself.

---

What no AI coding tool can do

• Understand your users — that is User Research’s job
• Make your MoSCoW decisions — that requires your team’s judgment
• Replace your Sprint Goal with a working feature
• Give you confidence to attempt something beyond your understanding — if you cannot follow what it built, the scope was wrong

---

Which tool for which situation

Situation	Recommended tool
On campus / XJTLU network	Copilot (VS Code), Windsurf, or Qwen Code
Not a coder — need to build something visual	TRAE SOLO mode (describe what you want, it builds)
Building a UI — want to point at the problem	Cursor 3.0 Design Mode (off-campus) or TRAE
Comparing two approaches to the same problem	Cursor 3.0 /best-of-n (off-campus)
Large refactor across many files	Cursor 3.0 agent, Windsurf agent, or Copilot agent mode
Open-ended task (“build me a backend”)	Break it into smaller tasks first, then either tool
Explaining or debugging code	Any — Claude.ai / DeepSeek work well for this
No IDE, just need to think through a problem	Claude.ai, DeepSeek, or Kimi (browser-based)

If you are a non-coder on the team

Your role is User Researcher, Product Manager, or Designer — not developer. You are not expected to write code. If your team needs UI scaffolding and no developer is available, TRAE SOLO mode is designed for this: describe what you want visually, and the agent builds a prototype you can test with users. You still need to understand what it built well enough to explain it at Checkpoint.

If a tool is blocked in China

• Copilot — works via the VS Code extension on most campus networks. This is your primary tool if you are on campus.
• Qwen Code — Alibaba-hosted, works everywhere in mainland China, 1,000 free requests per day.
• Windsurf — Codeium-hosted; accessible from most Chinese networks. Test on your connection — if it works, it is the strongest campus alternative to Cursor.
• TRAE — ByteDance-hosted, reliable from campus. SOLO mode requires no coding knowledge.
• Cursor — does not work from mainland China. U.S.-hosted models are restricted. Off-campus with international network access: it works and the student Pro year is worth claiming.
• DeepSeek / Kimi — domestic alternatives, reliable from campus, browser-based.