Making AI Reliable in the SDLC: The Role of Deterministic Context at Scale
summary
Claims concrete outcomes: grounding LLM coding agents in CAST’s deterministic, graph-based context cuts token use about 50%, doubles result accuracy and compresses a 7+ year mainframe modernization to 3 years, proven on a half‑million‑line Java app and a 4 million‑line case where two architects produce assessments in four weeks. Implementation uses CAST Imaging/Highlight, MCP integration into VS Code, Neo4j graphs, Agentic Suite/Kiuwan/AWS Transform to generate spec files, APIs and test suites and restore TDD; measurements report reduced token cost, higher correctness and repeatable impact analysis.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:09AI context in brownfield apps
- 2:53AI shortcomings in analysis
- 4:20Token efficiency in discovery
- 6:28Architecture risks with AI
- 7:11Token reduction with CAST
- 9:59CAST reverse engineering
- 12:49Demo: AI with CAST in VS Code
- 15:16Legacy transformation framework
- 17:43Factory workflow for modernization
- 21:16Agents with CAST data
- 23:25Human-readable specs mapping
- 26:34CAST Highlight for portfolio analysis
- 29:24Mainframe modernization outcomes
- 31:53Key artifacts for client planning
- 33:38Project metrics and results
- 34:44Q&A
related talks
28mWhy Software Quality Is the Key to Scaling AI Development
46mEnhancing Agent Reliability Through Evaluation-Driven Development
15mUsing AI effectively in large Java codebases
25mFrom Chaos to Control: Building Enterprise-Grade LLM Agents That Actually Work
29mBuilding Trust in AI Agents
31mBuild Intelligent AI Agents With Context Engineering
47mCode Is Cheap. Software Isn’t.
57mVerification, Cost, and Speed: What's Next for AI Coding at Scale