Good-bye page

Thank You for Completing the Course!

Thank you for your dedication in working through this material. You've learned something more durable than any product tutorial: the mental models, workflows, and disciplines of AI-assisted development — direction, context, verification, and security — that transfer to whatever tool you use next.

📌 The whole course in one breath

Treat the AI as a fast, literal, un-accountable junior engineer. Direct it well (rules, context, prompts, specs) and verify it well (review, tests, security). Learn the category and the standards, not one product's buttons.

🧭 Keep learning — organized by durable concept

We deliberately lead with canonical, maintained sources — specs, official docs, and standards bodies — over individual creators. These are still useful in two years.

Project context & standards

• agents.md — the open, cross-tool standard for project instruction files (spec + examples).
• modelcontextprotocol.io · github.com/modelcontextprotocol — the MCP spec, reference servers, and SDKs.

Prompt & context engineering

• Anthropic — Prompt engineering
• OpenAI — Prompt engineering
• Google — Gemini prompting strategies

(Read more than one — you're learning the shared principles, not one vendor's quirks.)

Tools (official docs — canonical, not YouTube)

• Claude Code docs · Cursor docs · OpenAI Codex · Aider

Spec-driven development

• GitHub Spec Kit — one open, model-agnostic toolkit for spec-first workflows.

Community example banks (useful, but tool-specific — label them as such)

• Awesome Cursor Rules — community .cursor/rules examples. Handy inspiration; remember your portable base is AGENTS.md.
• Awesome MCP Servers — maintained catalog of MCP servers (apply the trust caution before installing any).

Opinion / news (optional — clearly one creator's take)

• A couple of creator channels (e.g. Fireship for fast explainers, Theo for deep dives) are fine for staying current — just treat them as opinion, not canonical reference.

📊 How to tell what's actually good — benchmarks & your own evals

You'll constantly face "which model/tool is best right now?" Answer it with data, not marketing:

• SWE-bench (Verified) — an execution-based coding benchmark: does the model's patch make real tests pass? A durable comparison method, not a fixed ranking.
• ⚠️ Caveat: benchmarks are signal, not gospel. Leaderboards can be gamed or contaminated by training data — in Feb 2026 OpenAI publicly stepped back from SWE-bench Verified over exactly this. Use them as one input.
• 🧪 Build your own tiny eval set — 5–20 real tasks from your codebase with known-good outcomes, run candidates against them. Nothing predicts performance on your work like your work. (See Choosing Models & Tools.)

🎯 Don't memorize a ranking that expires next month — keep the method: public benchmark + your own evals.

Last reviewed: 2026-07. This resource list is a maintained artifact — links and tools drift; the concepts they map to don't. If you spot something stale, that's feedback we'd love.

Happy — and careful — building. 👋