full dune
Online(Remote)
Topic: Building a Governed Multi-Agent AI Coding Environment
The Concept
Moving beyond simple chat prompts to a structured Agentic Workflow that treats AI agents like a professional dev team. This session demonstrates how to orchestrate multiple LLMs to build, test, and document production-grade Systems/Rust projects autonomously.
The Workflow
The Builder (Antigravity): The primary code writer. It operates under strict .agents/rules (Rust standards, project specs) to ensure architectural integrity from line one.
The Orchestrator (TODO.md): Our "Source of Truth." It maintains the project state and ensures the user and agents stay perfectly synced on current objectives.
The Specialist (Gemini CLI): Triggered automatically on Git Commit/Push. It analyzes the code diff to generate Unit Tests and update project logs.
The Quality Gate: SonarQube and linting act as a hard gateway. If the code fails to meet "Global Engineering Standards," the sub-agents are automatically triggered to refactor the debt.
The Innovation: "Living Memory"
We eliminate Documentation Rot by using a dedicated docs/ architecture:
Feature Reports: Markdown files generated per feature that detail the "how" and "why" of the implementation.
Workflow Logs: Automated reports on the agent's decision-making process.
Post-Push Sync: By tying Gemini CLI to the Git lifecycle, every push results in a perfectly synced doc report and CHANGELOG.
Why it Matters
This solves the biggest hurdle in AI coding: Context Drift. By maintaining a structured, human-readable documentation hub, the project maintains a "Living Memory" that any developer (or another AI) can jump into and understand instantly.
balmy lantern