AI agents hit overengineering backlash

Frameworks like LangChain and CrewAI are increasingly criticized for "abstraction soup" that complicates debugging and increases token costs. Multi-agent systems often introduce non-determinism and infinite loops that solve problems a basic script with if-statements could handle more effectively. Sophisticated memory and RAG setups are frequently overkill for tasks that only require a well-structured system prompt or small-file context. Senior engineers are shifting toward "agent-lite" patterns, using direct LLM API calls and Pydantic for production-grade maintainability. This backlash marks a shift toward reliability and simplicity as developers move past the initial prototyping phase into production.