What are the main AI agent orchestration patterns?
The patterns most guides now use come from Anthropic: prompt chaining, routing, parallelisation, orchestrator-workers and evaluator-optimizer, plus the fully autonomous agent loop. For systems with several agents, OpenAI describes two more: a manager agent that calls others as tools, and decentralised hand-offs between peers. Each pattern trades predictability for flexibility to a different degree.
- Published
- 22 October 2026
- Sources checked
- 22 October 2026
- From
- Kramiva
01
The patterns
Summarised from Anthropic's “Building effective agents” and OpenAI's practical guide.
- Prompt chaining: a fixed sequence where each model call works on the previous output, with optional checks between steps
- Routing: classify the input first, then send it to a specialised prompt, model or process
- Parallelisation: run independent subtasks at once (sectioning), or run the same task several times and compare (voting)
- Orchestrator-workers: a central model breaks the task down at run time and delegates the pieces to workers
- Evaluator-optimizer: one model produces, another critiques, and the loop repeats until the result passes
- Autonomous agent: the model plans and uses tools in a loop until it decides the goal is met
- Manager (OpenAI): one agent keeps control and calls specialist agents as tools
- Decentralised hand-off (OpenAI): an agent passes the whole task to a peer, which takes over
02
How frameworks express them
Google's Agent Development Kit ships Sequential, Loop and Parallel workflow agents, with graph-based workflows added in ADK 2.0. LangGraph represents any of these patterns as a graph of nodes over shared state. The OpenAI Agents SDK offers two primitives that map onto the multi-agent patterns: agents used as tools (manager) and hand-offs (decentralised).
03
An illustrative example
A document-review system might route each upload by type, chain extraction and summary steps for invoices, and use an evaluator-optimizer loop for contract summaries, where a second model checks that every obligation in the source appears in the summary. Only one part of the system is a loop; the rest is fixed.
04
Trade-offs
The patterns toward the top of the list are easier to test and cheaper to run. Orchestrator-workers and autonomous agents handle more variety but are harder to predict and debug. Anthropic's recommendation is to combine the simple patterns first and move down the list only when measured results justify it.
Where Kramiva fits
Kramiva's AI automation and AI product feature work includes this design stage: which steps are fixed, which are model-driven, and where a person approves before anything irreversible happens.
Sources
- Building effective agents — Anthropic, 19 December 2024
- A practical guide to building agents — OpenAI
- Agent Development Kit (ADK) — Google
- LangGraph overview — LangChain
- OpenAI Agents SDK documentation — OpenAI
Accessed 22 October 2026. Kramiva is not affiliated with these publishers.
Related questions
Tell us what you are building.
A short brief gets you a real reply from a founder within one business day — an honest read on fit, scope, and budget.