One Week with Opus 5.5 (and Jev) - THIS is literally next level!
A comprehensive overview of Claude Opus 5.5's capabilities demonstrated through week-long testing, featuring comparisons with other AI models and practical applications including interactive 3D environments, game development, and productivity system visualization. The speakers highlight Opus 5.5's superior speed, efficiency, and token cost-effectiveness compared to previous versions.
Summary
This livestream presents extensive testing and findings from one week of using Claude Opus 5.5 alongside a new reasoning model called Jefferson (Jeff). The presenters compare Opus 5.5 against previous models like Claude 4.5, GPT-6, and demonstrate its capabilities through multiple practical experiments.
Key findings show Opus 5.5 is significantly faster, more accurate, and more token-efficient than Fable (Claude 3). The testing reveals that while Opus 5.5 performs exceptionally as a standalone model, the combination of Opus 5.5 with Jefferson creates particularly powerful synergies. Jefferson excels at making real-time decisions in low-latency scenarios, while Opus 5.5 handles planning and complex reasoning. This partnership is demonstrated through games where AI plays against itself, with Jefferson making frame-by-frame decisions and Opus 5.5 planning strategy.
The presenters showcase a creative tool called the PKA folder—a knowledge system containing AI agents that can collaborate across different sessions and even machines on a network. Using Opus 5.5, they've created interactive 3D environments including a virtual house representing their productivity system, allowing visual navigation through their information architecture. They also generated photorealistic 3D assets using Blender without specialized 3D skills, demonstrating how Opus 5.5 understands complex domain knowledge.
They discuss the importance of having a solid foundation—clear methodology, well-structured information, and domain expertise—before leveraging AI for maximum results. The cost comparison shows Opus 5.5 is 40% more cost-effective than Fable while being superior in capabilities. A critical observation is that token limits have shifted from being a limiting factor with Fable to allowing continuous productive use at a fixed price with Opus 5.5.
The speakers emphasize that successful AI implementation requires proper system design, not just access to powerful models. They highlight how their team uses AI agents that communicate and coordinate with each other in real-time, removing friction from execution workflows and shifting human focus toward decision-making rather than implementation.
Key Insights
- Opus 5.5 completely replaced Fable for the speaker's use case due to being much faster, more accurate, and consuming significantly fewer tokens while maintaining superior capabilities
- Jefferson (the new reasoning model) and Opus 5.5 work optimally in complementary roles: Jefferson makes fast real-time decisions at low latency, while Opus 5.5 handles planning and complex reasoning, demonstrating that the combination is more powerful than either alone
- The quality of results from AI models depends more on how well you structure information and explain your situation than on the power of the model itself—having a clear foundation and methodology matters more than raw model capability
- With Opus 5.5's improved token efficiency and pricing, the speaker can now operate productively all week while only reaching 50% of weekly token limits, eliminating the friction and limiting concerns that existed with Fable's constraints
- Using AI agents that coordinate with each other and can operate across different machines on a network creates a fundamental shift in workflow where humans become decision-makers rather than executors, increasing productivity and satisfaction
Topics
Transcript
[0:02] Hello everyone. We don't have time because there is so much to consider today. This is madness. Uh, one week into OPOS 5.5, and this live stream will be all about all the different experiments , findings, conclusions, tips and tricks, maybe even from Paco and me when it comes to this. But in short, it's madness. Paco, do you confirm? Absolutely. Yes. Absolutely. Absolutely. We haven't talked to each other in detail about this yet, but I'm [0:32] sure we've come to the same conclusions, and I'll start this presentation by hopefully sharing that right away. I've prepared a slide deck, or my team has , and that will wrap up all the testing that I've done that…
Full transcript available for MurmurCast members
Sign Up to AccessMore from The AI Productivity Podcast
Claude Opus 5.5 replaced Fable for us. Here is the proof.
The speakers demonstrate how Claude Opus 5.5 has replaced their previous AI model (Fable) for development work, showcasing significant improvements in speed, cost efficiency, and token consumption. They present multiple real-world applications including game AI, 3D rendering, design systems, and productivity tools built on their iCore methodology, emphasizing that proper system foundation and information structuring matters more than raw model capability.
myPKA Is Not Obsidian. Neither Is ICOR for Life.
Tom and Parker clarify that myPKA is a concept for building AI-assisted productivity systems using a local folder architecture, not a tool. They explain the three-layer system structure (AI LLM layer, local folder layer, and front-end interface layer) and why Obsidian was chosen as the recommended front-end while maintaining tool-agnosticism. The ICOR methodology provides the underlying framework regardless of which tools are used.
Obsidian Is the Surface. Here Is Where myPKA Fits.
The speakers clarify the architecture of their myPKA (Personal Knowledge Assistance) system, explaining how it consists of three independent layers: AI LLM services (Claude, ChatGPT), a local folder structure with markdown files, and a front-end interface (such as Obsidian). They emphasize that Obsidian is merely one optional tool for navigating the underlying system, not a replacement or lock-in, and demonstrate how the same core methodology can be implemented with different front-ends.
Obsidian is just a shell. Your knowledge folder is the real system.
Tom and Paco discuss how Obsidian is merely a shell for implementing the tool-agnostic iCore methodology, emphasizing that the underlying knowledge folder structure and methodology matter far more than any specific application. They introduce iCore for Life, a free Obsidian scaffold with custom plugins designed to help users understand and implement the productivity system's concepts and workflows.
Stop copying our AI setup. Copy how we think.
Tom and Paco discuss their different approaches to building AI-integrated productivity systems, emphasizing that the methodology (IICOR) should be implemented flexibly based on individual thinking styles and workflows rather than copied exactly. They showcase various visualization tools for managing deliverables, team knowledge, and work streams while stressing the importance of documentation and local folder structures over relying on external databases.