Code Is Cheap. Software Isn’t.
summary
AI generates boilerplate and whole methods in seconds but teams hit failures when code touches millions of lines of business logic, implicit invariants, or multi‑year long running transactions; verification catches hallucinations. Recommended stack: use IBM Bob or Claude with MCPs like Quarkus Dev MCP or DocMCP/Context7 to ground agents, keep changes small (module‑level), require JUnit and integration tests, use local Git rollbacks and ADRs, decompose work, treat generated code as a pull request, and phase explore → review → fix.
AI-written summary · about
chapters (17)
- 0:00Introduction and keynote
- 2:02The initial magic of AI
- 3:00Challenges in real codebases
- 4:16Implicit knowledge and failures
- 7:35AI hallucinations and context
- 9:48Strategy for AI engineering
- 19:34Decomposition and best practices
- 21:29The Monday morning rule
- 22:14Context and workflow engineering
- 25:32Managing architecture and data
- 27:44Compression of information
- 28:18Commands and skills
- 33:05Tooling and MCP servers
- 37:18Reviewing generated code
- 40:15Cognitive load and burnout
- 41:01The Java developer advantage
- 45:14Conclusion and takeaways
related talks
19mCode Is Cheap. Software Isn't
32mEscape the Try-Again Loop: Verifying AI-Generated Code
46mCode Quality in the AI era, needed more than ever
57mVerification, Cost, and Speed: What's Next for AI Coding at Scale
28mWhy Software Quality Is the Key to Scaling AI Development
44mAI Code Generation - Benefits, Risks and Mitigation Controls
40mCode Rot at Machine Speed: Preventing AI-Induced Technical Debt
18mCoding Reinvented: How AI Agents Are Shaping the Future of Software Development