An LLM is not an agent: the 10 building blocks of agentic systems, from orchestration to RAG
goyalshaliniuk · x · 2026-11-19
The author argues an LLM alone is not an AI agent — it becomes agentic when the surrounding system can understand goals, plan, use tools, remember context, act, and recover from failure. The 10 building blocks:
- Foundation model: reasoning, generation, understanding, tool calling
- Agent orchestration: task routing, workflows, state, retries
- Context & instructions: goals, rules, requirements, tools
- Memory & state: conversation history, preferences, progress
- Knowledge & RAG: trusted retrieval and grounding
- Tools & actions: APIs, databases, code execution
- Planning & workflow: decomposing goals, sequencing, tracking
- Multi-agent collaboration: delegation and handoff between specialized agents
A generic framework explainer without novel detail.
More from coding & agent
- OpenCode fix cuts GPT-6.2 uncached input per step from ~14,000 tokens to 3 — victor_explore · 2026-11-19
- Remotion ships real motion blur, addable via AI agents with one prompt — Vjeux · 2026-11-19
- SREGym open-sources a benchmark replaying real incidents like Cloudflare and Reddit outages for SRE agents — tianyin_xu · 2026-11-19
- Uber, SpaceX and Nvidia engineers pack first AI Software Factory summit PortCon — Pavan_Belagatti · 2026-11-19
- Dev builds James Bond-style stealth game with GPT 6.2 Astra — chongdashu · 2026-11-19
- Why autonomous agents outgrow MCP and collapse back to APIs + scoped tokens — uriwa · 2026-11-19