This Day in AI Podcast Nano Banana 2 is Here! Gemini-3 Shutdown & The AI Layoff Myth | EP99.36
266 snips
Feb 27, 2026 They test Google’s Nano Banana 2 for cheaper, faster image generation and push its limits with annotation-based edits and slide fixes. They dig into why Gemini-3 was pulled and how agentic workflows expose model flaws. The conversation compares Opus, Codex and GLM-5, explores smart model routing and cost trade-offs, and questions whether recent mass layoffs are really about AI.
AI Snips
Chapters
Transcript
Episode notes
Nano Banana 2 Halves Image Costs And Enables Fast Iteration
- Nano Banana 2 cuts image costs ~50% and targets speed by using the Flash model for faster inference.
- Michael Yo notes the price drop makes iterative workflows (like slide generation) far more practical and cheaper per-deck.
Annotate Images To Fix The Last 10 Percent
- Annotation-based editing dramatically improves targeted image edits and accuracy.
- Michael Yo and Chris Sharkey show circling/marking elements then prompting leads to precise edits like removing an element or adding a portrait exactly where marked.
Last Mile Design Kills Perceived Quality
- The hard 'last mile' on design is small defects that shatter perceived quality despite strong 95% results.
- Chris Sharkey explains replacing a wrong logo or tiny diagram error turns a polished deck into 'amateurish' without precise layer-aware edits.