YouTube digest 2026-06-01 (4 worth watching)

[JustinPerea]

Bookmark Queue Append

https://www.youtube.com/watch?v=NuePCNMpWGc
https://www.youtube.com/watch?v=N7b1PJc7SFc
https://www.youtube.com/watch?v=UQKg0td-Bf4
https://www.youtube.com/watch?v=pM40Nv0LZrs

Copy block above into ~/Documents/Obsidian Vault/Bookmark Queue.md, then run /process-queue.

Preview

Can LLMs generate Enterprise Quality Code? — Prasenjit Sarkar, Sonar — AI Engineer

TL;DR: Sonar benchmarked 53 LLMs on 4,444 Java assignments — Claude Sonnet 4.6 has the highest security-issue rate (300/M LOC), and they propose an ACDC pipeline (guide → verify → solve) with pre-commit SonarQube in 1–5s.

Engineering voice agents: Latency, quality, and scale — Rishabh Bhargava, Together AI — AI Engineer

TL;DR: Concrete latency budget for voice agents — users hang up above 1s; LLM dominates (200–300ms TTFT, 8–30B param sweet spot); colocating STT+LLM+TTS drops 75ms cross-DC overhead to ~5ms.

Spec-Driven Testing for Agents With A Brain the Size of A Planet — Steven Willmott, SafeIntelligence — AI Engineer

TL;DR: Bigger models aren't automatically safer — they understand jailbreaks (e.g., malicious instruction wrapped in a poem) that small models ignore; agents need explicit specs covering rules, ontologies, roles, and robustness — not just test datasets.

Full digest

See daily/2026-06-01-youtube.md for all 11 videos (4 watch / 7 skim / 0 skip), full insights, and tools mentioned.

---

Note: transcripts were unavailable from this environment (YouTube blocked both youtube-transcript-api and yt-dlp). Summaries are based on RSS-feed titles + descriptions only — the AI Engineer talks ship with detailed abstracts that made this workable, but Shorts have ~no description text and were judged on title alone.

---

Generated by Claude Code