---
type: source-digest
topic: agent-wiki
status: active
created: 2026-04-06
updated: 2026-04-06
source_notes:
  - "[[clippings/26-02-26  anthropic.com  Building Effective AI Agents]]"
derived_from:
  - workspace-initialize
confidence: high
source_url: https://www.anthropic.com/engineering/building-effective-agents
---

# anthropic-building-effective-ai-agents

## Why This Matters

This source adds a strong product and engineering constraint to the workspace: start with the simplest approach that can work, then add workflows or agents only when that simplicity stops being enough.

## Key Takeaways

- Agentic systems should be chosen deliberately rather than assumed by default.
- Workflows and agents are different; the former follow predefined code paths, while the latter let the model direct its own process.
- Developers should begin with direct API usage or simple compositions before introducing framework abstraction.
- Complexity trades off against latency, cost, predictability, and debuggability.
- Reliable tools and clear tool interfaces are foundational for useful agent behavior.

## Entities

- Anthropic
- workflows
- agents
- augmented LLM
- tool interfaces

## Related Concepts

- [[research/agent-wiki/concepts/simplicity-before-autonomy|Simplicity Before Autonomy]]
- [[research/agent-wiki/concepts/compiled-knowledge-layer|Compiled Knowledge Layer]]

## Open Questions

- Which parts of the agent-wiki workflow should stay coded and deterministic versus model-directed?
- Where should this vault prefer workflows over more autonomous agent behavior?
- What counts as "too much abstraction" inside a knowledge-maintenance system?