Builders ditch LangChain: why more devs roll their own agent harness
raftconsensus · reddit · 2026-10-20
A Reddit discussion on the "build vs. wrap" spectrum for agent workflows: many builders get frustrated by framework lock-in and context bloat in LangChain/LangGraph or multi-agent frameworks, and end up scripting custom harnesses. Three discussion axes: (1) failure modes that trigger building your own — state recovery, tool sandboxing, deterministic lint/test gates, API token bleeding; (2) actual stacks — building on LangGraph/CrewAI, wrapping Claude/Codex/Gemini CLI, or a bare-metal Python/TS loop with standard APIs; (3) where it gets messy — context window management, runaway tool loops, over-engineered state machines. Useful firsthand picks for engineers choosing agent architecture.
More from coding & agent
- Eval researcher teases guide to subjective-task evals, with a curated resource list — cwolferesearch · 2026-10-20
- On building persistent agents: owning your agent's memory stack — omarsar0 · 2026-10-20
- Mintlify launches Autopilot beta: an always-on agent that keeps docs in sync with your code — ycombinator · 2026-10-20
- 10 GitHub repos that feel like leaked future: Rust Photoshop, SSD-streamed LLM inference — Shruti_0810 · 2026-10-20
- Google's T2J-Bench shows coding agents score just 0.676 migrating PyTorch repos to JAX — CShorten30 · 2026-10-20
- MCP tool-poisoning detector broken in a day—here's the context-aware fix and what it still misses — blitzcrieg11 · 2026-10-20