https://datatweets.com/courses/ai-agents/multi-agent-systems/Recent content in Multi-Agent Systems on DATATWEETSHugoenCopyright (c) 2025 DatatweetsSun, 28 Jun 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/multi-agent-systems/lesson-1-why-and-when-multiple-agents/Fri, 30 Jan 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/multi-agent-systems/lesson-1-why-and-when-multiple-agents/Piling every job onto one agent bloats its prompt and blurs its focus. Splitting work across specialists helps — at the cost of coordination. Learn the real trade-off, why to exhaust a single agent first, and the three multi-agent patterns this module builds: agents-as-tools, routing, and orchestrator-workers.https://datatweets.com/courses/ai-agents/multi-agent-systems/lesson-2-agents-as-tools/Fri, 30 Jan 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/multi-agent-systems/lesson-2-agents-as-tools/A specialist agent is exposed to a supervisor as a plain tool: the supervisor calls research_destination(question) like any tool, and behind it a full sub-agent runs its own run_agent loop with its own retrieval tool. You build it in stages, watch a retrieved fact propagate from the sub-agent all the way up into the supervisor’s final answer, and see why context isolation makes this pattern so clean.https://datatweets.com/courses/ai-agents/multi-agent-systems/lesson-3-routing/Fri, 30 Jan 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/multi-agent-systems/lesson-3-routing/Routing sends each request to exactly one specialist. A lightweight classification call reads the request, picks a label, and dispatches to the matching handler — cheaper and sharper than one agent handling every category. You’ll build route(), add a fallback so a bad label never crashes the system, and see it verified against an SDK-shaped mock.https://datatweets.com/courses/ai-agents/multi-agent-systems/lesson-4-orchestrator-workers/Fri, 30 Jan 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/multi-agent-systems/lesson-4-orchestrator-workers/Routing picks one specialist; orchestrator-workers runs many and combines them. Build an orchestrator that decomposes a goal into subtasks, runs a worker per subtask, and synthesizes one plan. Learn why independent workers can run in parallel, and when a job is really sequential plan-then-execute instead.https://datatweets.com/courses/ai-agents/multi-agent-systems/lesson-5-guided-project-a-multi-agent-atlas/Fri, 30 Jan 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/multi-agent-systems/lesson-5-guided-project-a-multi-agent-atlas/Put the module together on Atlas. Build a small team — a retrieval-grounded researcher, a budget analyst, and an itinerary writer — then wire in all three patterns: the researcher as a tool, a router that dispatches by category, and an orchestrator that decomposes a trip into subtasks and synthesizes one grounded, cited plan.