Framework · 2023–now · AI Present
LLM orchestration frameworks
Chains, tools, memory, and agents as a framework — LangChain-class glue sold as architecture. Fatigue was real; the repos became MCP clients, not tombstones.
How LLM orchestration frameworks mutated
LLM orchestration frameworks faded as defaults when the abstraction was heavier than the product. They mutated into adapters: first-class MCP, thinner SDKs around vendor tool APIs, and fewer wrappers around the same chat completion. What sticks is structured tool use, evals, and a protocol; what faded is orchestration as the product.
Cost of the fad
What LLM orchestration frameworks cost
Teams adopted chain/agent frameworks before they had a single reliable tool call. Abstraction layers multiplied while prompt quality stayed flat.
Case studies
Patterns
Compare with
Related
Framework · 2023–now
Multi-agent orchestration hype
LangGraph, CrewAI, AutoGPT cosplay — then vendor harnesses shipped the same loop as an API. Open-source swarms faded; leased orchestration stuck.
$ Autonomous agent swarms burned API budgets on coordination loops that a single prompt and a cron job would have handled. Debugging "which agent lied" became the new on-call sport.
Framework · 2023–now
LLM app frameworks
LangChain-class glue mutated into MCP clients and thin wrappers around vendor APIs. The durable pieces are still boring: evals, retrieval, and product UX.
Practice · 2023–now
RAG as default architecture
Retrieval-augmented generation as the answer to every knowledge problem — then "just stuff the window" as the counter-fad. Often right; often Postgres with pgvector would suffice.
$ Teams stood up vector pipelines, chunking strategies, and rerankers before asking if fine-tuning or a SQL query would answer the question. Retrieval infra became the product.
Practice · 2023–now
Agent ops / LLM observability
Tracing, cost caps, and prompt versioning for production LLM features — mostly constrained tool loops, not autonomous agents. Datadog for tokens.
Architecture · 2024–now
Context / tool protocols
Standard ways for models to talk to tools and data — MCP after the 2026 stateless rewrite, not one-off plugin snowflakes.
Platform · 2025–now
Vendor agent APIs
The Codex/Claude loop as a product SKU — durable sessions, compaction, subagents, and a sandbox you do not own.
$ Teams replaced a weekend of LangGraph glue with a managed session API, then discovered residency, recovery, and kill switches all live on the vendor side of the harness.