Vibe Coding & Agentic AI: Key Takeaways from ChaiyoGCP & Build with AI Bangkok 2026

AI-assisted development, agentic systems, GenUI, and event reflections.

Also on Medium

How AI is reshaping developers — from writing code to designing intelligent systems

Buit with AI

ก้าวข้ามขีดจำกัดการพัฒนาซอฟต์แวร์ด้วย AI Agent และ GenUI

Attending two major tech events in a single day is exhausting but incredibly rewarding. On March 22, 2026, I had the chance to attend the ChaiyoGCP Season 6 Kick-off in the morning and Build with AI Bangkok in the afternoon.

การเข้าร่วมงาน Tech 2 งานใหญ่ในวันเดียวอาจจะเหนื่อยหน่อย แต่สิ่งที่ได้รับคุ้มค่ามาก เมื่อวันที่ 22 มีนาคม 2026 ที่ผ่านมา ภัทรได้มีโอกาสเข้าร่วมงาน ChaiyoGCP Season 6 Kick-off ในช่วงเช้า และต่อด้วย Build with AI Bangkok ในช่วงบ่าย นี่คือสรุปประเด็นสำคัญที่ได้เก็บตกมาฝากกันค่ะ

This post is not a full workshop recap, but rather a curated highlight of key insights that changed how I think about AI, development, and the future of software engineering

บทความนี้ไม่ได้สรุป workshop ทั้งหมด แต่เป็น insight สำคัญที่เปลี่ยนมุมมองของภัทรเกี่ยวกับ AI และการพัฒนา software


1. ChaiyoGCP Kick-off: The Rise of the AI Developer

Getting Started with Google Antigravity

By Jirayut Nimsaeng

The idea of “vibe coding” was introduced — but not in a lazy way.

It’s about structured AI-assisted development, not just prompting randomly.

The concept of "Vibe Coding" isn't just about fun; it’s about structured planning. การทำ "Vibe Coding" ไม่ใช่แค่การเขียนโค้ดตามความรู้สึก แต่คือการวางแผนที่เป็นระบบ:

  • AI development still requires engineering discipline:
    • Planning
    • Rules & Standards: We need clear code generation and style guides.
    • SKILL.md: A new way to define skills for AI, including Code Review, Unit Testing, and QA.
    • The AI Developer: Developers aren't disappearing. Instead, we are evolving into "AI Developers" who must accelerate growth (Junior to Senior) within just one year.

AI compresses the learning curve, but only if you structure your workflow properly

Evaluating AI Agents with Vertex AI

By Rewat Airamaneerat

Context is king. If an agent lacks context, it fails. บริบท (Context) คือหัวใจสำคัญ หาก Agent ขาดบริบทที่ถูกต้อง ก็ไม่สามารถทำงานได้จริง:

  • Agent Architecture: AI Agent -> MCP (Model Context Protocol) -> LLM.
  • Evaluation is Key: Using Vertex AI to grade agents via:
    • Matching strategies:

      • Exact match (rule-based)
      • In-order match (workflow validation)
      • Any-order match
      • Single-tool usage
    • Metrics:

      • Precision / Recall
      • Semantic evaluation (meaning-based)
      • LLM-as-a-judge
      • RAG metrics

AI evaluation is becoming a first-class engineering problem, not just experimentation.



2. Build with AI: Moving from "Chatting" to "Doing"

Vibe Coding & Rapid Prototyping

By Saad Hamid

The workflow has shifted: Think -> Build -> Publish. กระบวนการทำงานเปลี่ยนไปเป็น: คิด -> สร้าง -> เผยแพร่

  • Tools: Using AI Studio for prototypes, Antigravity for agentic IDEs, and Gemini CLI for terminal smarts.
  • Powerful Prompting: Don't be lazy. Define roles, goals, and create a "vibe" to get the best results.
    • Good prompts outperform lazy prompts:
      • Assign a role
      • Define clear goals
      • Decide if AI is needed
      • Set the “vibe” (context)
      • Optional: include visuals

Designing Agentic AI Systems in Google ADK

By Joan Santoso

We are moving from Assistants (Task-oriented) to Agents (Goal-oriented). เรากำลังเปลี่ยนจากผู้ช่วย (Assistant) ที่ทำตามสั่ง เป็น Agent ที่มุ่งเน้นผลลัพธ์ (Goal-oriented)

  • Design Patterns:
    • Sequential: Step-by-step pipeline.
    • Parallel
    • Review & Critique: One agent generates, another reviews.
    • Coordinator dynamic routing
    • Human-in-the-loop: Vital for sensitive cases; humans approve before the final response.
  • Evolution: Start simple and evolve as workloads or AI capabilities grow

ADK Capabilities

  • Multi-agent systems (sub-agents)
  • State management
  • Callback systems (e.g., safety checks)

Phases of AI System Design

  1. Assessment

    • Task complexity
    • Performance needs
    • Budget
    • Human involvement
  2. Decision

    • Single-agent vs Multi-agent
  3. Evolution

    • Systems must adapt:
      • Workload changes
      • Business needs evolve
      • New AI capabilities emerge

AI architecture is never final — it evolves continuously

Dynamic UIs with GenUI and Gemini

By Amorn Apichattanakul

The future of UI is Agentic UI. อนาคตของ UI คือ "Agentic UI" หรือ UI ที่ปรับเปลี่ยนตามสถานการณ์

  • A2UI Protocol: Supports React, Angular, Flutter, and Lit (with SwiftUI/Jetpack Compose coming Q2 2026).
  • Static screens are becoming Adaptive UX—personalized and dynamic interfaces generated in real-time by AI.

Frontend is shifting from designing screens → designing experiences powered by AI


Personal Reflection / มุมมองส่วนตัว

This day made one thing clear: The barrier to building has dropped, but the bar for "Architectural Thinking" has risen. In the past, we spent 80% of our time writing syntax. Now, AI handles the syntax, but we must spend that 80% on System Design, Context Management, and Evaluation. If you can "vibe" with the AI and guide it with the right architecture, you can build products at a speed that was impossible two years ago.

สิ่งที่ภัทรเห็นได้ชัดจากทั้งสองงานคือ "กำแพงในการสร้าง (Building) นั้นต่ำลง แต่มาตรฐานของการคิดเชิงสถาปัตยกรรม (Architectural Thinking) นั้นสูงขึ้น"

ในอดีตเราอาจใช้เวลา 80% ไปกับการเขียน Syntax แต่ตอนนี้ AI จัดการตรงนั้นให้เราแล้ว เราจึงต้องเอา 80% นั้นมาใช้ในการ ออกแบบระบบ, จัดการบริบท (Context), และการวัดผล (Evaluation) แทน ถ้าคุณสามารถ "Vibe" ไปกับ AI และชี้นำมันด้วย Architecture ที่ถูกต้อง คุณจะสร้าง Product ได้เร็วในระดับที่เมื่อสองปีที่แล้วยังทำไม่ได้เลยค่ะ

And honestly…

That’s both exciting and terrifying.


Resources & Credits


tags:

  • GDGCloudBKK
  • ChaiyoGCP
  • BuildWithAI
  • GoogleCloud
  • GenAI
  • VibeCoding

Written by Pat (Napatchol) | Full-Stack Developer