Agentic Spectrum
The spectrum of AI agent architectures from simple prompt-response to fully autonomous multi-agent systems, with trade-offs at each level.
Last updated
After this section you can
- Tell a workflow from an agent by who picks the next step
- Match a task to prompt chaining, routing, parallelisation, orchestrator-workers, evaluator-optimiser or an agent
- Justify the lowest level that works in terms of calls, latency, failure modes and the cost of a mistake
09
The Agentic Spectrum: Workflows to Autonomy
Not every LLM feature needs an agent. Seven building blocks, from one augmented call to a model running its own loop, and the discipline of choosing the lowest one that works.
Key idea
In a workflow, your code decides the path and the model fills in the steps. In an agent, the model decides the path. Every move toward the agent end buys flexibility with calls, latency and predictability, so start with the simplest block that solves the task.
Seven building blocks, one dividing line: who picks the next step, your code or the model