Reusable multi-agent patterns: prompts, tools, context
Code-level multi-agent patterns on Claude: role-scoped prompts, per-agent tool scoping, typed handoff contracts, need-to-know context, and critic wrappers.
8 min read
Agentic AI & LLMs
Agentic AI & LLMs archive page 306 of 631
Agentic AI & LLMs
Agentic AI, LLM engineering, and the models behind modern automation — multi-agent systems, LLM evaluation and comparisons, RAG, fine-tuning, AI infrastructure, security, and production AI engineering.
From the blog
Latest articles
Build a multi-agent system on Claude: a walkthrough
Step-by-step guide to building an orchestrator–subagent system on Claude: decomposition, parallel dispatch, compact handoff, looping, and synthesis.
7 min read
Build an AI Triage Agent for Your SOC: A Walkthrough
Build a Claude-powered security triage agent step by step: schemas, MCP tools, the run loop, approvals, and the evals you need before you trust it.
7 min read
AI-Accelerated Offense: Security Defense Architecture
Architect a Claude-based security program that matches AI-accelerated attackers — ingestion, context, reasoning, action, and governance planes end to end.
7 min read
Multi-agent coordination patterns: the architecture inside
How multi-agent systems are wired with Claude: orchestrators, subagents as context firewalls, state transport, and the five coordination patterns that matter.
8 min read
Continued Pretraining in LLMs: From Foundation to Domain Intelligence
Continued Pretraining in LLMs: From Foundation to Domain Intelligence
2 min read
AutoGen 0.5: Async Core, Extensions Model, and What Changed
Microsoft AutoGen 0.5 reworks the core to async-first and ships the extensions model. The architectural shift, the migration path, and what to do about it now.
10 min read
LangGraph Subgraphs vs Supervisor Pattern: Which to Pick in 2026
Subgraphs and supervisor patterns both compose multi-agent systems in LangGraph. The trade-offs that matter once you scale past three agents.
11 min read
SGLang vs vLLM: 2026 Throughput Benchmarks on Real Workloads
SGLang and vLLM are the two serious open-inference servers in 2026. Head-to-head benchmarks on Llama, DeepSeek, and Qwen workloads with reproducible methodology.
11 min read