ERO Product Docs

Appearance

Thoughts expand_moreAgentic Project Management is ​

导出时间: 22/07/2026, 19:58:38

Thoughts ​

Agentic Project Management is the core operational methodology used to scale the Event Route Optimiser (ERO) platform from a local set of scripts into an enterprise-grade, multi-cloud data product[1]. Rather than executing development ad-hoc, this framework establishes a highly structured, traceable, and secure relationship between human product owners and autonomous AI agents[1][2].

The model is defined by several key architectural pillars:

Core Pillars ​

1. The Human-Agent Team Hierarchy ​

Under this governance model, tasks are divided among specialized human and AI roles to maintain architectural control[3]:

  • The Product Owner (Human): You act as the lead strategist, defining commercial objectives, priorities, and validation criteria[3].
  • The System Architect (Gemini): Serves as the high-level brain, designing data schemas, resolving integration logic, and drafting the strict technical specifications and instructions for the workspace[3].
  • The Pair Programmer (GitHub Copilot): Operates directly inside the editor (VS Code) to handle rapid boilerplate generation, real-time code completion, and syntactical debugging[3].
  • The Agentic Orchestrator (Antigravity IDE / AGY): Executes autonomous, multi-agent development workflows[4]. Agents monitor repositories, run integration tests, manage package dependencies, and handle infrastructure deployment[3].

2. Traceable BDD Backlogs (The Guardrails) ​

To prevent autonomous agents from hallucinating or drifting in scope, they are given strict, deterministic boundaries[1]:

  • Azure DevOps (ADO) as the Engine Room: Standard engineering execution and sprint planning are managed inside a single ADO project (EventRouteOptimiser), keeping intellectual property and roadmap tracking secure[5][6].
  • Behavior-Driven Development (BDD) User Stories: Backlog items are authored with strict Given/When/Then Acceptance Criteria[5]. For example, mapping exactly what happens when the Cloudflare Worker receives an API call[5][7].
  • Deterministic Limits: The agents are legally bound to only generate or mutate code that directly satisfies these BDD criteria, ensuring full security and auditability[1][5].

3. Autonomous Backlog Generation via MCP ​

To bridge the gap between local design documents and cloud-based project management, the agents leverage the Model Context Protocol (MCP)[8]:

  • The ADO MCP Server: Using a secure Personal Access Token (PAT) locked in a local, git-ignored .env.local file, the agent establishes a bridge to your Azure DevOps organization[9][10].
  • One-Click Backlog Publishing: The agent autonomously parses structured local Markdown specifications (such as docs/04-ado-backlog.md) and calls the ADO API[11][12]. It instantly generates the entire work hierarchy—mapping Epics to Features, User Stories, and Tasks with full parent-child relationships[11][13].

4. Idempotency and Workspace Autonomy ​

To make developer onboarding completely frictionless, the execution of the project setup must be entirely automated and safe[14]:

  • Idempotent Execution: Master instruction logs (like SYSTEM_BUILDER.md) are designed so that the agent always audits the current workspace before running an installation or applying a change[15][16]. If a configuration or dependency is already present, it is marked as verified and skipped[16].
  • Allowed Command Overrides: To prevent high-friction workflows where you have to constantly click "Allow" for every terminal command, VS Code's workspace settings are updated with auto-approve command patterns[17][18].
  • Safe Wildcards: Granting permission specifically to CLI prefixes like pwsh*, npm*, npx*, git*, and terraform* allows the agent to handle branch creation, dependency compilation, and infrastructure planning autonomously, without giving it permissions to run destructive commands[19][20].

By utilizing this Docs-as-Code strategy, the entire system specification lives alongside the codebase[21]. Whether you are working on your laptop or your desktop, the agents can instantly absorb the system's state, sync with your active Azure DevOps board, and execute the next sprint safely[16].

🚀 Since we mapped out the transition from local development to Azure subscription environments, I can draft a Bicep parameter template for your test and pre-production Resource Groups

引用来源 ​

[1] AI-Powered Mobile App Development Blueprint [2] AI-Powered Mobile App Development Blueprint [3] AI-Powered Mobile App Development Blueprint [4] AI-Powered Mobile App Development Blueprint [5] 1. Engineering & Delivery (Your ADO Sweet Spot)Az… [6] AI-Powered Mobile App Development Blueprint [7] AI-Powered Mobile App Development Blueprint [8] AI-Powered Mobile App Development Blueprint [9] AI-Powered Mobile App Development Blueprint [10] AI-Powered Mobile App Development Blueprint [11] AI-Powered Mobile App Development Blueprint [12] AI-Powered Mobile App Development Blueprint [13] AI-Powered Mobile App Development Blueprint [14] AI-Powered Mobile App Development Blueprint [15] AI-Powered Mobile App Development Blueprint [16] AI-Powered Mobile App Development Blueprint [17] AI-Powered Mobile App Development Blueprint [18] AI-Powered Mobile App Development Blueprint [19] AI-Powered Mobile App Development Blueprint [20] AI-Powered Mobile App Development Blueprint [21] AI-Powered Mobile App Development Blueprint