AI and Agents

Kendamaryu noted that AI agents need to be cloud-based to be easily introduced to non-engineers, implying that local agent setups present a usability barrier for broader adoption.^[1] They also explored the concept of implementing "collective intelligence" by using LLMs to simulate multiple agent personalities, drawing a parallel to how prediction markets like Polymarket often outperform individual human analysts.^[2]

Regarding Claude's Fable 5 model, kendamaryu reported that restrictions were lifted and the model was being restored to users on a rolling basis starting the next day.^[3] They later shared that Claude AI announced Fable 5's return,^[4] and that Fable 5 subscription access was extended until July 12.^[5] In a personal note, they expressed a dilemma between enjoying a July 4th three-day weekend and using Fable 5.^[6]

Kendamaryu observed that Claude's Artifacts feature had become a regular tool, particularly for creating UI mockups.^[7] They also described using Gemini's Embedding 2 model to store vector embeddings of files in a database, enabling semantic search across text, images, video, and audio. They suggested that Claude Code can be instructed to add this embedding database and create a search skill to leverage it.^[8] The sudden release of Embedding 2 surprised them, as it came right after the previous version.^[9]

On a more critical note, kendamaryu admitted that while experimenting with AI is engaging and fun, a "so what?" feeling sometimes overwhelms them, questioning the deeper purpose of the activity.^[10] They also reported a usage issue with the Max plan where performing a single question from a 0% usage state immediately jumped to 15% consumption, suspecting a systemic problem affecting the entire window.^[11]