Spec > Spell: Write It Clearly, and Let AI Do the Magic

A session from LINE THAILAND DEVELOPER CONFERENCE 2025 by Pathai Suwannawiwat, Solution Engineer, LINE Thailand

Transcript

1. 𝗦 𝗽 𝗲 𝗰 > 𝗦 𝗽 𝗲 𝗹 𝗹 : 𝗪 𝗿 𝗶 𝘁 𝗲 𝗜 𝘁 𝗖 𝗹 𝗲 𝗮 𝗿 𝗹 𝘆 , 𝗮 𝗻 𝗱 𝗟 𝗲 𝘁 𝗔 𝗜 𝗗 𝗼 𝘁 𝗵 𝗲 𝗠 𝗮 𝗴 𝗶 𝗰
2. • Then & Now • Vibe coding is spell • Engineering the Spellbook • From spells to specs • Conclusion Agenda
3. Then & Now From past limits to present breakthroughs, AI is no longer tomorrow it is today!
4. Just 315 Days Everything have changed …
5. LLM Late 2022 Agentic AI Early 2023 MCP Mid 2024 LLMs in coding are here.
6. LLM + AgenticAI + Getting LLMs to help with coding is no longer far away. But prompting them isn’t casting spells. LLMs in coding are here.
7. Vibe coding is spell Vibe coding feels like magic, but leaves systems chaotic.
8. "Building software with LLMs without carefully examining the code it writes" - Simon Willison What is Vibe Coding?
9. Vibe coding with LLMs is like writing code with no practices to guide it. Systems become fragile and hard to trust when vibes take over. That’s why a software engineering mindset matters. The Cost of Vibe Coding
10. We do more than just code. Through planning, design, testing, and refactoring, we focus on solutions that make our systems stronger and more efficient. Software Engineering Mindset
11. When engineering mindset meets good LLM prompting LLMs are more than tools. They can be trusted teammates. With clear instructions and the right mindset, working with AI feels like working with a real colleague.
12. Engineering the Spellbook The building blocks of context engineering for reliable AI collaboration.
13. What is good LLM prompting? Everyone knows what good engineering mindset looks like. To apply it with LLMs, we need context engineering. Let’s see the core elements that make LLMs cast spells for us.
14. Prompt Engineering VS Context Engineering Context Engineering Prompt Engineering Prompt Engineering It is subset of Context Engineering using for crafting the best instruction for a specific task. Context engineering means giving LLMs extra information, like system background, code style, and rules, so they can handle many tasks better.
15. Available Tools: External resources or APIs (e.g., MCP, databases, search). i.e. MCP it’s The Chef's "Secret Master Technique" Context Engineering Components Context System Prompt User Prompt Available Tools - Philipp Schmid - System Prompt: Defines the role, rules, and boundaries of the AI. It'is like Chef’s Rulebook. It tells what to do and what not to do. User Prompt: The direct instruction & order for individual task. It’s is Your Order to the chef.
16. System Prompt: In VSCode, we can create a file at /.github/copilot- instructions.md This file works like our system prompt. User Prompt: It’s a prompt which you give to LLMs to perform any task.
17. Available Tools MCP Client: VSCode is MCP client then we can install server by add config to /.vscode/mcp.json
18. From Spells to Specs Bringing mindset, tools, and prompts together into clear specs.
19. OAPlus is a large microservice system, and each service uses a complex architecture such as hexagonal.
20. Step 1 Prepare the available tools for LLMs
21. Engineering Mindset in Action Gohan CLI is our internal tool that helps teams create new services instantly from a project starter template. - Embeds best practices from the start - Makes setup faster and smoother - Reduces errors and avoids rework - Keeps every project aligned with the same standards
22. Empowering Gohan CLI with MCP Server 1. Gohan CLI 2. MCP Server 3. LLMs 4. Result
23. Install Gohan MCP Server
24. Step 2 Embedding Context into System Prompts
25. Prepare an effective System Prompt
26. Step 3 Guiding with Clear User Prompts
27. Provide User Prompt
28. Summary 1. System Prompt
 (copilot-instruction.md) 2. User Prompt
 (Your order to LLMs) 3. LLMs are now your teammates 🫶.
29. Conclusion
30. “Maintaining job security is your responsibility.”
31. Thank you