Lesson 3: AI Pair Programming Patterns

The Problem with Unstructured AI Help

When you ask AI to "add a new feature" without context:

• AI doesn't know your architecture constraints
• AI might suggest changes that break dependencies
• AI doesn't understand your team's conventions

Architecture-Aware AI Interactions

Pattern 1: Before Changing X, Ask About Dependencies

You: "What would break if I change UserService?"

AI (with Sruja context): "Changing UserService would affect:
- LoginComponent (depends on UserService)
- ProfileUI (calls UserService.getProfile())
- NotificationService (receives user events)

You should also update:
- user-auth contract (API boundary)
- test/UserService.test.ts (affected tests)
"

Pattern 2: Use AI to Explore, Then Validate

# AI helps explore
"What does the checkout flow look like?"

# AI shows understanding
"Based on your architecture, checkout involves:
- CartService -> PaymentService -> OrderService

Shall I trace through each step?"

Then validate with Sruja:

# After AI proposes changes
sruja lint
sruja drift -r .

Pattern 3: AI-Assisted Impact Analysis

# Before making a big change
"Use sruja impact to show me blast radius of removing the legacy import service"

# AI calls MCP tool
# Blast radius analysis shown

Pattern 4: Ask for Sruja DSL, Then Validate

You: "Generate Sruja DSL for a new notification service"

AI: [generates Sruja code]

You: "Validate this with sruja lint and check it fits our patterns"

Pattern 5: AI as Architecture Reviewer

You: "Review this code change for architectural compliance"

AI with context: [checks against policies, finds issues]
"1. This change violates your 'no-direct-db-access' policy
2. New dependency creates circular reference with BillingService
3. Missing required 'tech' field in container definition"

Pattern 6: Ask the host to query lineage (MCP)

When the Sruja MCP server is connected, the model can pull in-repo traces without guessing:

• sruja_get_context_events — recent rows from .sruja/context_events.jsonl (after intent check, drift, or a merged proposal).
• sruja_get_agent_learnings — AgenticMemory entries tied to an element id (same idea as “memory hits” in sruja focus JSON).

Use these before risky edits: they surface prior guardrails and verification history the same way a human would scan Slack—except here it is structured and local to the repo. Full parameter list: mcp_tools_reference.md.

Effective Prompts for Architecture Context

Workflow: AI-Assisted Feature Development

1. Define the task

   "I need to add a new payment method to checkout"
   

2. AI explores context

   AI: "Let me check your PaymentContract and checkout flow..."
   

3. AI proposes with constraints

   AI: "Based on your architecture, this requires:
   - Adding endpoint to PaymentContract
   - Updating FraudDetection rules
   - New scenario test for the payment flow"
   

4. Human approves

   "Sounds right, implement it"
   

5. AI implements

6. Human validates with Sruja

   sruja lint
   sruja validate --scenarios
   

Anti-Patterns to Avoid

Hands-On: Practice AI Pair Programming

1. Set up context for your task:

   # Focus on the area you're working on
   sruja focus --file src/payment.rs -r .
   

2. Ask architecture-aware questions:

   "What does this component depend on?"
   "What would break if I change this?"
   "Show me the blast radius of this change"
   

3. Validate AI suggestions:

   # After AI proposes changes
   sruja lint repo.sruja
   sruja drift -r . -a repo.sruja
   

4. Use impact analysis:

   # Check what would be affected
   sruja impact ComponentName -r . --depth 3
   

5. Optional — with MCP, ask for lineage before a large change:

   • sruja_get_agent_learnings for your target element id
   • sruja_get_context_events with a small limit to see recent intent/drift/proposal events

Learning Outcomes

• ✅ Apply architecture-aware AI interaction patterns
• ✅ Use AI to explore and understand architecture
• ✅ Validate AI suggestions with Sruja commands
• ✅ Use MCP lineage tools (sruja_get_context_events, sruja_get_agent_learnings) when available
• ✅ Avoid common anti-patterns in AI-assisted development

Quiz: Test Your Understanding

Q1: What should you do before making a significant change?

A) Just make the change quickly B) Ask about dependencies and use impact analysis C) Delete the old code first D) Ignore architecture concerns

Q2: Why should you validate AI suggestions with Sruja?

A) Because AI is always wrong B) Because AI can hallucinate and make mistakes C) To make the code faster D) It's not necessary

Q3: What command shows the blast radius of changing a component?

A) sruja lint B) sruja context C) sruja impact D) sruja drift

Q4: Which MCP tools surface in-repo lineage for an AI session?

A) sruja_get_repomap and sruja_bm25_search B) sruja_get_context_events and sruja_get_agent_learnings C) sruja_add_element and sruja_commit_evolution D) sruja_validate_change only

Module Complete!

You've completed the AI Editor Integration module. You now understand:

• ✅ Setting up Sruja MCP server
• ✅ Building architecture context for AI
• ✅ Using context-driven AI assistance (including git-range focus and on-disk lineage)
• ✅ Effective AI pair programming patterns (including MCP lineage queries)

This module completes the Agentic AI course with practical AI editor integration skills.