From simple swarm to a living idea ecosystem.

Build the core ant-like system.

• ·Three agent types: Generator, Evaluator, Combiner
• ·Three pheromones: Novelty, Evidence, Constraint Fit
• ·Three actions: Create, Score, Combine
• ·Shared notebook as the intelligence layer
• ·Local context only: top notes, random notes, recent notes
• ·Evaporation and reinforcement
• ·One synthesis step at the end

Run 50–100 blind benchmark prompts.

• ·Constraint Fit
• ·Evidence
• ·Novelty
• ·Organizing Principle
• ·Decision Value
• ·Overall Preference

Expected strengths: strategy, positioning, creative synthesis, framework discovery, decision framing. Expected weaknesses: debugging, technical runbooks, strict procedural tasks.

Focus the product around the strongest signal. Colony should become a tool for finding:

• ·Hidden patterns
• ·Governing principles
• ·Strategy frameworks
• ·Creative concepts
• ·Decision models

Allow ideas to survive across runs. The notebook becomes long-term memory.

• ·Winning notes persist
• ·Losing notes decay
• ·Reused frameworks gain strength
• ·Failed patterns lose strength
• ·Idea lineages become visible

Let benchmark outcomes adjust the swarm.

• ·Tune pheromone weights
• ·Tune exploration / exploitation ratio
• ·Tune agent prompts
• ·Retire weak behaviors
• ·Promote successful behaviors

Move beyond judge scores. Track whether recommendations actually worked.

• ·Did the GTM strategy produce signups?
• ·Did the positioning improve conversion?
• ·Did the framework help decision-making?
• ·Did users reuse the output?

The long-term vision is not a bigger chatbot. It is a living idea ecosystem.

Ideas should appear, compete, combine, mutate, survive, and disappear. The system becomes valuable because it evolves concepts over time.