New blog posts, comparisons

[amavashev]

New blog post: docs/blog/cycles-vs-llm-proxies-and-observability-tools.md (~2,700 words)

The post covers:

1. Opening scenario — a concrete weekend overspend story where LiteLLM and Langfuse both worked perfectly but couldn't prevent $2,800 in damage
2. Three layers framework — Routing (proxies) vs Visibility (observability) vs Authority (Cycles)
3. LLM Proxies section — what LiteLLM/Portkey do well, where they stop (only see model calls, report after, no atomic reservations, no hierarchical scopes), comparison table, and how to use both together
4. Observability section — what Helicone/Langfuse do well, where they stop (post-hoc, alert latency math, no enforcement), comparison table, and how to use both together
5. Full production stack — architecture diagram showing all five layers
6. Next steps — 6 cross-links to existing docs and blog posts

Cross-links added from two concept pages:

• from-observability-to-enforcement-*.md → new blog post
• how-cycles-compares-to-rate-limiters-*.md → new blog post

The background Plan agent completed with a more detailed outline. Comparing its suggestions against the blog post I already wrote and pushed, the Plan agent suggested a few additional elements worth considering:

1. A master comparison table naming all four tools (LiteLLM, Portkey, Helicone, Langfuse) individually across 14 capabilities
2. A "When you need which layer" decision framework section
3. Human latency cost table at multiple agent speeds (I included this but with slightly different numbers)