The AI Native Dev - from Copilot today to AI Native Software Development tomorrow What Developers Can Build Next With AI
24 snips
Nov 12, 2025 
Join Baruch Sadogursky, an expert in specification-driven development, as he discusses the importance of compiling human-readable specs into trustworthy tests. Liran Tal from Snyk delves into the risks of relying on LLM-suggested security fixes, highlighting real-world vulnerabilities. Alex Gavrilescu, author of BacklogMD, explains minimal markdown tasks for AI agents to prevent 'vibe coding.' Lastly, Josh Long from Broadcom showcases Spring AI integrations for Java applications, emphasizing AI's seamless connection to existing business logic.
AI Snips
Chapters
Transcript
Episode notes
Lock Down Critical Flows With Tests
- Prioritize writing specs and acceptance criteria for areas you care about; allow LLMs freedom only on low-risk parts.
- Increase test coverage on critical flows so generated code can't deviate from desired behavior unnoticed.
Protect Tests Before Letting Models Write Code
- Protect compiled tests (make files read-only or containerize them) so generated code cannot alter acceptance checks.
- Then allow agents to iterate on code until the protected tests pass, ensuring correct behavior.
AI Helps Draft Specs But Humans Must Verify
- AI can generate specs from high-level prompts but hallucination and human ambiguity mean specs need human review and iteration.
- Use multiple models to generate and validate specs, then iterate until the readable spec matches intent.