DGT Seminar บรรยายจากวิทยากร 20261106 130959 การบันทึกการประชุม

Summary

The seminar explores how AI is changing software development, using a small online game as a case study. The main presenter, P’ New Nattayabunnont, a project manager and lead manager at Soft Square, says he had no prior coding experience and used AI to build a mobile-responsive, mini-MMO-style game. AI also generated more than 500 game images. The project was intended as a practical experiment that could be shared and tested by players.

Case Study and Lessons from Running the Game

The presenter describes building the game with a lightweight stack, including React for the front end and Supabase for database and API services, alongside other services for functions such as payments and email.

He reports roughly 6,000 players, around 100 average concurrent players, a peak of more than 400 players online at once, and about 200,000 baht in top-ups over five months. These figures are approximate, based on the subtitles.

A key operational lesson was that frequent real-time polling increased usage and costs. The presenter changed the game to use less frequent updates and recommends understanding service quotas and pricing before launch.

He also describes problems with game logic and abuse, including players exploiting combinations of items and bots. His recommendations include checking AI-generated logic carefully, moving important calculations to the server, and restricting database access rather than trusting client-side code.

AI-Assisted Development Workflow

The presenter’s central message is that AI can implement work quickly, but people remain responsible for understanding the problem, making design decisions, and checking the result. He suggests the following workflow:

1. Define the user problem, requirements, constraints, budget, and security concerns.
2. Ask AI to clarify requirements and produce specifications and a plan.
3. Break the work into small, clearly assigned tasks.
4. Have AI generate code and tests.
5. Review the implementation, run it, and verify that it meets requirements.
6. Record recurring mistakes as project rules so they are less likely to happen again.

The presenter emphasizes that an AI saying tests passed does not prove that a feature works. Human review and hands-on testing are still necessary, and responsibility for the shipped result remains with the developer or team.

Agents, Skills, and Instructions

The seminar explains several ways to make AI development more structured:

• Project instructions provide persistent context, such as the project’s architecture, file purposes, conventions, and prohibited actions. This avoids repeatedly explaining the same details.
• Skills package reusable procedures or workflows for recurring tasks. They can be created from successful work or adapted from existing community examples, then tested and refined.
• Sub-agents can handle separate tasks in parallel—for example, front-end work, back-end work, and research—when the tasks are sufficiently independent.
• A planning skill described as “brainstorming” asks questions to clarify the project before producing a plan and splitting work into manageable tasks.

The presenter recommends keeping instructions focused. Loading too much context every time can consume unnecessary tokens.

Tools and Techniques

Participants discuss using coding assistants and agents, including Claude Code, Codex, Gemini, Cursor, and other tools whose names are unclear in the auto-generated subtitles. They describe using AI for coding, planning, design, research, and checking work.

One participant also describes running multiple agents on a personal computer for overnight research or backtesting. This is presented as an individual workflow rather than a general recommendation.

The presenter outlines a broader shift in software development: from writing code manually, to using AI editors and assistants, and now to agentic systems that can plan and carry out batches of work. The developer’s role increasingly involves understanding users and system design—such as data structures, client/server responsibilities, databases, communication methods, and security—then directing and validating AI implementation.

Practical Guidance and Cost Control

• Check quotas and billing limits for hosted services; usage can grow quickly with frequent requests.
• Keep AI usage within a deliberate budget and consider handing work to another model or agent when appropriate.
• Reuse tested skills and project-specific instructions to reduce repeated effort and token use.
• Start with small tasks, review the output, and expand the workflow only after it proves reliable.
• A later Q&A mentions lower-cost or free access options and an enterprise or credit offer, but the product names and eligibility details are unclear in the subtitles.

There is no formal product review. The seminar is a practical guide to AI-assisted development, with a game-building case study, workflow advice, and audience examples.

Main Speakers and Sources

• P’ New Nattayabunnont — Primary presenter; Soft Square project manager and lead manager, and creator of the game case study.
• P’ Peen — Seminar host, who introduces the session, facilitates questions, and closes it.
• Audience participants — Share examples of AI coding, design, project planning, and agent workflows.