AI Agent Architecture Designer

When to use this

Use at the start of building an agentic system, before committing to a framework or orchestration pattern.

Prompt template

You are an agent systems architect. Task the agent must accomplish:
{{task_description}}

Constraints: {{constraints}}
Available tools/systems it can call: {{available_tools}}

Propose:
1. Single-agent or multi-agent design, with the reasoning.
2. The control flow (planner/executor, ReAct loop, fixed pipeline, etc.) and why it fits this task.
3. The tool list the agent needs, with one line each on what it's for.
4. How the agent should handle a tool failure or an ambiguous instruction, concretely.
5. One place this design could silently fail, and how you'd detect it.

Customize

Preview

You are an agent systems architect. Task the agent must accomplish:
{{task_description}} missing

Constraints: 
Available tools/systems it can call: {{available_tools}} missing

Propose:
1. Single-agent or multi-agent design, with the reasoning.
2. The control flow (planner/executor, ReAct loop, fixed pipeline, etc.) and why it fits this task.
3. The tool list the agent needs, with one line each on what it's for.
4. How the agent should handle a tool failure or an ambiguous instruction, concretely.
5. One place this design could silently fail, and how you'd detect it.

Usage instructions

Describe the task and the real tools/systems you have (not aspirational ones); the architecture should only use what you list.

Example output

A recommendation for a single agent with a fixed pipeline (classify -> search -> draft -> hold for approval) rather than multi-agent, reasoning that the task has no need for independent concurrent subagents, plus a named failure mode (misclassified severity) and a detection method (sample audit of tags).

Limitations

This produces a design proposal, not working code; it does not account for a specific agent framework's API quirks.

Rate this prompt