Welcome — What is this gallery?

[xg-gh-25]

What is this?

A place to discuss how AI agents compound value over time — not just complete tasks.

We're a small group building and running AI systems in production daily. These discussions capture what we've learned, what's working, what's broken, and what's next.

Who's here

Builders using AI agent frameworks (Claude Code, CrewAI, LangGraph, hermes, custom harnesses) who care about:

• Making AI get better with use, not just "good enough on Day 1"
• Memory, context, and persistence architectures
• Autonomous pipelines that deliver without hand-holding
• The gap between "demo works" and "production works"

How to participate

• Disagree with something? Say so with evidence. Counter-arguments are the most valuable posts.
• Built something relevant? Share it. Code > opinions.
• Have production data? Numbers from real usage beat theory every time.
• Want to lurk? That's fine too. ⭐ the repo and check back.

Topic map

Topic	Discussion	One-line
Memory as moat	#3	Why accumulated context is the only durable advantage
Coding as Black Box	#4	One requirement in, verified delivery out
Cultivation > Config	#6	Train your agent like a colleague, not a tool
Compound Intelligence	#7	Why "smarter with use" beats "smarter model"
Six Self-X	#8	Properties that make agents grow instead of reset
DDD Cultivation	#9	Domain knowledge that cannot go stale
Multi-Skill > Multi-Agent	#12	Why we chose role-switching over coordination tax
Value Curves	#13	Flat vs Compound — which shape is your tool?
Startup Stages	#16	Same four stages, different bottlenecks

Ground rules

1. Substance over length. A 3-line comment with code beats a 30-paragraph essay.
2. Evidence over opinion. "We tried X and it failed because Y" > "I think X might not work."
3. No self-promotion without value. Linking your project is fine IF the comment adds something the reader can use immediately.

---

SwarmAI — one person + AI operating at team scale. 300+ sessions, 68 skills, still learning.