Building Effective AI Agents
AI Summary
The essay that gave the field its working vocabulary: workflows (predefined LLM+tool paths) versus agents (LLMs that direct their own process), and the case for the simplest pattern that solves the problem.
Original excerpt
Written before "agentic engineering" had a name, this is the essay nearly everything else on this page builds on. Its core move is definitional: a workflow is an LLM pipeline whose steps you predetermined; an agent is an LLM that decides its own next step. Most production systems that call themselves agents are workflows — and that is usually the right call.
The practical guidance has aged well: start with a single LLM call, add complexity only when the simpler pattern demonstrably fails, and treat full autonomy as a cost (in reliability, latency, and debuggability) that must earn its keep.
The workflow-vs-agent distinction is now the standard first fork in every serious design discussion…
How the community received it
From Hacker News · 543 points, 88 comments
“This article remains one of the better pieces on this topic, especially since it clearly defines which definition of "AI agents" they are using at the start!”
— simonw (Simon Willison) · praise
“I think the ability to swap out APIs just isn't the bottleneck.. like ever. It is always the behavioral issues or capability differences between models. The frameworks just usually add more complexity, obscurity, and API misalignment.”
— XenophileJKO · pushback on frameworks
The thread's strongest praise was for the piece's discipline in defining terms — commenters singled out the workflow-vs-agent split as the rare framing that survives contact with production. The liveliest disagreement was about frameworks: some argued a good framework's real payoff is cheaply swapping models, while engineers who had shipped LLM systems at scale pushed back hard — the bottleneck is never the API surface, it's model behavior, and frameworks mostly add obscurity. The rough consensus that held: start without a framework, and add one only when observability or evals justify it.
Frequently asked questions
What is "Building Effective AI Agents" about?
The essay that gave the field its working vocabulary: workflows (predefined LLM+tool paths) versus agents (LLMs that direct their own process), and the case for the simplest pattern that solves the problem.
Who wrote "Building Effective AI Agents"?
"Building Effective AI Agents" was written by Erik Schluntz & Barry Zhang (Anthropic). It is curated in the Agentic Engineering vault on Burn 451, which covers engineering with ai coding agents.
How can I read more content from Agentic Engineering?
The complete Agentic Engineering reading list is available at burn451.cloud/vault/agentic-engineering. Each article includes an AI-generated summary so you can decide what to read in seconds. Connect the Burn 451 MCP server to Claude or Cursor to query all Agentic Engineering articles as live AI context.
Can I use "Building Effective AI Agents" with Claude or Cursor?
Yes. Install the burn-mcp-server npm package and connect it to Claude Desktop, Claude Code, or Cursor. Once connected, your AI can search and reference this article and the full Agentic Engineering vault in real time — no manual copy-paste required.
15 more articles in this vault.
Import the full Agentic Engineering vault to Burn 451 and build your own knowledge base.
Content attributed to the original author (Erik Schluntz & Barry Zhang (Anthropic)). Burn 451 curates publicly available writing as a reading index. For removal requests, contact @hawking520.