Software Engineering Challenges in the AI Coding Era
1. Core Challenges of the AI Coding Era
No matter how far AI Coding evolves, there are always three dimensions of scope in software development that require human definition:
| Scope Dimension | Question to Answer | Consequence if Ambiguous |
|---|---|---|
| Requirements Scope | What to build? What not to build? For whom? What defines success? | AI produces functionally correct but directionally wrong output |
| Technical Scope | Which tech stack? How to layer? Where are module boundaries? | AI-generated code is fragmented and cannot be integrated |
| Data Scope | What entities exist? What are the relationships? How does data flow? | Data silos, interface mismatches |
This aligns with the philosophy of Harness Engineering — not replacing human judgment, but "installing a steering wheel" on human domain judgment: let humans decide the direction, and let AI execute the driving. AI's execution power amplifies productivity, but it also amplifies the cost of scope errors. That is why scope definition matters more than ever in the AI era.
2. Our Answer: BA + SA Define Three-Dimensional Scope, PM Orchestrates Harness Engineering
The actual scope definition is handled by the BA and SA Agents — the former locks down business boundaries, the latter locks down technical boundaries:
Business Analyst"] SA["SA Agent
Solution Architect"] end subgraph SCOPE["Three Scope Dimensions"] D["Data Scope
Entities / Relationships / Flow"] R["Requirements Scope
What to Build / What Not
For Whom / Success Criteria"] T["Technical Scope
Tech Stack / Layering
Modules / Deployment / APIs"] end PM["PM Agent
Project Manager"] BA -->|"Produces Requirements Spec
Scenario Deep-Dive / Process Modeling
Data Dictionary / User Stories"| R BA -->|"Produces Data Dictionary
ER Diagrams / Field Specs
Data Flow Diagrams"| D SA -->|"Produces Architecture Spec
Overall Architecture / Module Decomposition
Data Architecture / Deployment Architecture"| T SA -->|"Produces API Docs
Deployment Guide"| T D -->|"Data Baseline"| PM R -->|"Requirements Baseline"| PM T -->|"Technical Baseline"| PM PM -->|"Harness Engineering
Iteration Planning & Scheduling
Workload Assessment / Progress Coordination"| OUT["Overall Project Plan
+ Demoable MVP"]
The BA Agent covers Requirements Scope and Data Scope, while the SA Agent covers Technical Scope. The PM Agent does not directly define scope, but acts as the "steering wheel" of Harness Engineering — based on the requirements baseline, data baseline, and technical baseline produced by BA/SA, it orchestrates iteration cadence, coordinates delivery progress, and aligns requirements with technology in the right sequence.
3. How Agents Truly Participate in the MVP Process
Taking Iteration 0 as an example, the collaboration flow of the three Agents is as follows:
- Phase 1 (Concept Input): PM collects materials and distills concept-level requirements; BA assists in understanding the business; SA assists in evaluating technical feasibility
- Phase 2 (Three-Dimensional Parallel): BA produces an initial PRD (scenario definitions, business processes, data entities, system boundaries) + MVP user story set; SA produces a technical risk assessment (risk list, preliminary architecture diagrams, integration plan) + development task list; PM integrates and produces the execution plan and project master schedule
- Phase 3 (Scope Confirmation): PM facilitates business-technical cross-review; BA and SA respond from their respective dimensions; finalize In/Out/TBD/Won't scope
- Phase 4 (Execution Period): Development team codes the MVP; BA deepens the full PRD (process modeling, data dictionary, full story set); SA deepens the full architecture (module decomposition, APIs, data, deployment, non-functional architecture); BA/SA intersperse reviews of development output
This four-phase model is the core framework for the PM Agent to drive Iteration 0. For detailed execution flow and task breakdown, see Iteration 0 Plan.
4. How to Get Started
You can explore in the following order:
- Install Guide — Connect Agents to your AI Coding tool
- BA Agent — Learn how to define requirements scope with Requirements Specifications
- SA Agent — Learn how to define technical scope with Architecture Design Specifications
- PM Agent — Learn how to orchestrate the big picture with iteration plans