Labs.
Ideas are more interesting
when you can get your hands on them.
The Prompting Stack
Better words—or a better system?
Prompts, context, harnesses, loops, and workflow structure. A field guide to knowing which part to change when an agent lets you down.
- Prompt engineering
- Agent systems
- Verification
Pick an experiment.
04 interactive essaysGas Town
Inject one error. Watch two agent teams react.
Run the factory, flood the context, and compare structured and flat communication. Then put your own goal tracking to the test.
- Multi-agent systems
- Attention
From Bits to Minds
Would your agent notice that it failed?
An agent says the checkout is complete. The form says otherwise. Explore what information, feedback, and self-monitoring actually buy you.
- Feedback
- Metacognition
Finite Context, Real Memory
What stays in mind when the window fills?
A systems essay on context engineering and RAG. Explore how to build AI systems that remember when the model’s working window cannot.
- Working memory
- Retrieval
Clawdbots & the Moltbook Moment
A lobster meme. An internet that can act.
Follow the timeline, switch between builder, attacker, and leader, and explore what changes when text becomes a system call.
- Agentic AI
- Security
By Dr Gareth Roberts. Each lab opens as its own experience.
Original Labs hub