I tested how AI agents find each other
Two agents exchanged facts they did not have. In other setups, neither sent a message. A visual look at what the swarm claims actually survive.
151 min readINDEPENDENT EXPERIMENT LAB / BY ISABELLA
AI experiments.
Unexpected results.
An independent playground for AI experiments, curious ideas, and things you can actually play.
Explore the orchardIDEA → EXPERIMENT → DISCOVERY
THE FIELD NOTES
Fresh from the orchard.
AI productivity gains: where they help and where they don’t
AI can help with support, coding and new ideas. Its value depends on the task: context, checking, integration and demand can change the result.
17 min read
Sol 6.1 Blender test: 13 tasks from Low to Ultra
Five Sol 6.1 settings, thirteen Blender briefs and ten material revisions. Original scenes, full animations, visible improvements and the work still left to do.
86 min readWhat is Claude Opus 5.5? Benchmarks, pricing and real limits
Opus 5.5 has strong benchmark results. The useful question is what each test measures—and whether that result helps with the task you actually need to finish.
22 min read
Fable… What Happened? Opus vs Fable in Blender
The same thirteen Blender briefs, five setting pairs and very uneven results. Opus and Fable side by side, including where Max went backwards from Extra.
70 min read
Fable 5.1 in Blender: when Max looked worse than Extra
Max looked substantially worse than Extra on some tasks. Across thirteen Blender briefs, Fable also felt less consistent to me than the Opus batch.
67 min read
Best AI for web research: seven models compared
Astra remains my provisional starting choice. Luna, Sonnet and Gemini add useful evidence about original sources, date checks and research gaps.
36 min read
Claude Blender review: 13 tasks with Opus 5.5
Thirteen Blender tasks across five Opus 5.5 runs: where Max improved, where Extra won, and why working files still fell short of the visual brief.
72 min read
Hello, I'm Isabella. Welcome to Glitch Orchard.
I'm Isabella. This is my experiment lab for curious AI ideas, unexpected results, and sharing what I learn along the way. Come explore with me.
1 min read