The Scale Saga: From Specs to 10x Engineering

A session at LINE THAILAND DEVELOPER CONFERENCE 2026

Transcript

1. The Scale Saga: From Specs to 10x Engineering AI Builds It. We Own the Outcome.
2. Agenda • Then & Now • Scaling Execution • The Human Bottleneck • Scaling Understanding • Conclusion
3. Then & Now Three Years. A New World.
4. 2024: Possibility What can we build with AI? Exploring agship models from OpenAI fl and product ideas
5. 2025: Uncertainty Where do engineers t? Context engineering and spec driven fi development
6. 2026: Responsibility We still own the outcome Understanding. Decisions. Care.
7. Scaling Execution From Intent to Impact.
8. Mission 10x A challenge to rethink productivity Redesign the fl work ow
9. And We Did We Integrated AI into Our Work ow. fl Planning Coding Review
10. Harness software that gives an AI model the context, Context tools, and work ow to get work done. Agent loop Rules LLM Examples: GitHub Copilot in agent mode, Claude Code, and fl Codex. Tools Feedback
11. 4 sprints Result estimated. Delivered in 2. From the old work ow to a new way of working across fl teams.
12. How We Made It Custom agents Work (i.e. Github Copilot for VS Code) # [agent name] ## Role Who are you? What is your job? ## Goal What should you achieve? ## Context What should you know or read rst? ## Rules What must you follow or avoid? (i.e. Team rules) ## Work ow What steps should you take? What should you do if you get stuck? *.agent.md ## Output What should you deliver? fi fl In what format?
13. Harness Engineering Harness Con gure Context Agent loop Rules .agent.md Create Review custom results. LLM Improve agent the agent. Finding for human review fi Engineer Tool Feedback
14. The Human Bottleneck Code Moves Fast. Can We Keep Up?
15. Inside the Agent Loop When Speed Outruns Understanding Agent Write code Debug Search These steps used to give us opportunities to learn
16. Research from Anthropic How AI Impacts Skill Formation — Shen & Tamkin / Anthropic, 2026
17. The Understanding Gap Average post-task quiz score 50% 67% With AI assistant Without AI assistant 17 percentage points Largest gap: debugging
18. AI Builds It. We Own the Outcome. We decide what ships and stand behind the result
19. Responsibilities Understand the change. Own the result
20. Scaling Understanding Turn Complexity into Clarity.
21. Engineering sense Keep human judgment active in the workflow Ask the Anticipate Know what right impact to verify
22. Step 1: Recall Before Generate Think first. Then compare with AI. Before AI answer After AI answer Recall Predict Compare How does the What needs to change? What matches? What system work What could go wrong? di ers? What did I miss? ff AI is not always perfect. Our judgment still matters.
23. Real use case: Where Does This Data Belong? Order Domain: Order domain Subscription Domain: Credit data Manages the subscriptions each account should receive based on orders. Subscriptio Receives purchase events and records orders. Credit Domain: Credit n domain domain Account Account Domain: domain Stores user details and tracks which subscriptions Tracks each account’s credit balance, grants, and usage.
24. Step 2: Visualize with HTML Make understanding visible
25. Demo 1 Architecture Change Map
26. Demo 2 UI & Database Simulation
27. Step 3: Explore Together A place to explore options. Make evolving ideas visible to the team Alternative idea Open questions Shared experiment
28. Team Space A place to explore options. Ask questions as a team. 01 Explore the same demo Make the behavior visible. 02 Compare the options Discuss bene ts and tradeo s. 03 Ask questions together fi ff Find gaps before we decide.
29. Step 4: Document Decisions From exploration to specification Requirements Decisions and rationale Business rules Acceptance criteria
30. fi OAPlus Doc From exploration to specification OAPlus Doc stores nal specs agreed on across teams after planning and brainstorming, ready for OAPlus Doc Built with Docusaurus development.
31. Summary Visual with HTML Document Approve changes Old Work ow New Work ow Recall before fl fl generate Explore together Harness
32. Conclusion AI Builds It. We Own the Outcome.
33. Responsibility The Outcome Is Still Ours Keep Learning. Keep Owning.
34. BUILD THE NEXT FRONTIER