Get started on Sui through Agentic Engineering
summary
Agentic engineering prescribes a concrete workflow: research, plan, implement where 80–85% of effort collects factual codebase context and a precise testing plan, then the agent writes code; frequent intentional compaction means dumping project state into Markdown and starting a fresh convo when the context window nears the 40–60% sweet spot (46% cited) to avoid context rot. The demo wires an agent to the Sui Docs MCP server so answers about kiosks, marketplaces and shared vs owned objects come from official docs, which cuts hallucinations and reduces AI-generated tech debt.
AI-written summary · about
chapters (25)
- 0:00Introduction
- 0:14Meet Raphael Chia from Mysten Labs
- 0:43Agentic Engineering on Sui
- 0:54AI Coding Productivity Research
- 1:34Goals for High-Quality AI Engineering
- 3:11Context Engineering Explained
- 4:34Why Prompting Feels Like a Casino
- 5:14Two Methods for Managing Context
- 5:29Frequent Intentional Compaction
- 6:48Writing Project State to Markdown
- 7:45How Tool Calls Consume Context
- 8:22Keeping Context Under Control
- 8:57Optimize for Quality, Not Token Savings
- 9:24Research, Plan, and Implement
- 10:06Research the Codebase First
- 10:45Create a Testable Implementation Plan
- 11:19Implement From the Research and Plan
- 12:14Adapt the Workflow to Your Project
- 12:34Prompt With Precise Requirements
- 13:35Building on Sui
- 13:48Using the Official Sui Docs MCP
- 14:33Live Documentation Demo
- 15:31Kiosks, Marketplaces, and Sui Objects
- 16:29Sui Quick Start Resources
- 17:03Closing Advice for Builders
related talks
27mAgentic Engineering: Working With AI, Not Just Using It
20mReimagining InnerSource for the Agentic AI Era
39mAgents for the Sake of Happiness
40mHow To Tame Your Agents?
40mAI Agents in Practice
49mHow to build with ASI tech stack
21mBest practices are trapped in your head and your AI has no idea
45mAI agents in practice - beyond the prototype