Part 2 · The Harness Around the Model

Agent: AI That Gets Things Done

Four capabilities: Plan / Tool / Memory / Act — click to see real-world cases

THE QUESTION THIS PAGE ANSWERS

ANSWER FIRST

What is the key idea behind “Agent: AI That Gets Things Done”?

Four capabilities: Plan / Tool / Memory / Act — click to see real-world cases

DECISION RULE

Make the claim earn its place. Use this page as a decision aid, not a definition to memorize. Connect the idea to one real task, one observable result, and one failure that would change your mind.

TRY NEXT

Write one question you could answer with evidence after trying this idea.

WATCH FOR

A conclusion that sounds complete but leaves the key assumption untested.

A regular LLM can only "say" things; an Agent can "do" them. The difference is that an Agent has four additional capabilities that allow it to execute tasks in the real world.
The Classic Architecture · Where It All Comes From
Lilian Weng's LLM Agent system overview diagram: the Agent at the center, connected to Planning, Memory, Tools, and Action
Lilian Weng's Twitter avatar
Lilian Weng@lilianweng
Former OpenAI VP of Research & co-founder of Thinking Machines Lab · Now back at OpenAI leading frontier research on recursive self-improvement (RSI)

Many of the ideas in this lesson — and in this entire Agent Engineering chapter — come from her June 2023 blog post "LLM Powered Autonomous Agents". A deeply respected pioneer, she laid out the profoundly influential Agent architecture in the famous "crab diagram" above: the Agent (LLM) sits at the center as the brain, reaching out to Planning, Memory, Tools, and Action.

This one diagram has shaped the entire AI Agent industry — nearly every Agent framework today traces back to it. We highly recommend following her on Twitter/X.

Four Core Capabilities (from Lilian Weng's Architecture) · Click to Switch Demo
🧭
Plan
Breaks complex tasks into executable step sequences — think before acting
🔧
Tool Use
Calls search, code execution, databases, and APIs — an extension of hands and feet
🧠
Memory
Short-term context + long-term vector storage — remembers users across conversations
🔄
Act / Reflect
Observes results after execution; on failure, automatically analyzes the cause and corrects course
Regular LLM vs Agent
Regular LLM
Question → Generate answer → Done
Can only "say," cannot "do"
Agent
Task → Plan → Call tools → Observe → Correct → Complete
Step-by-Step Demo
Plan — Full Competitive Analysis Workflow
Auto-plays on scroll into view · Click a capability card to switch demo
Takeaway
Takeaway Agent = LLM + Tools + Loop. The core is "Think → Act → Observe → Think again" — which is exactly Lilian Weng's architecture diagram in motion: the LLM is the brain, and Planning / Memory / Tools are its limbs. Every AI product is ultimately designed around this loop. The better the loop is designed, the more reliable the Agent becomes.

Why “The Classic Architecture · Where It All Comes From” can find relevant content

“Many of the ideas in this lesson — and in this entire Agent Engineering chapter — come from her June 2023 blog post "LLM Powered Autonomous Agents" .” moves retrieval beyond storing material: the real question is how to find what is relevant. That decision shapes the input quality of RAG, recommendation, and image-search systems.

Similarity is not the answer

In the flow described by “This one diagram has shaped the entire AI Agent industry — nearly every Agent framework today traces back to it.”, embeddings place items in a comparable semantic space and a neighbor index narrows the search. The final answer still depends on whether the retrieved chunks cover the question, whether the distance metric fits, and whether the evidence is current.

Separate findable from relevant

Turn “This one diagram has shaped the entire AI Agent industry — nearly every Agent framework today traces back to it.” into a small test: prepare queries with known answers, record relevance, misses, and distractors, then decide whether chunking, the index, or reranking needs to change.

From “The Classic Architecture · Where It All Comes From” to “Four Core Capabilities (from Lilian Weng's Architecture) · Click to Switch Demo”

“The Classic Architecture · Where It All Comes From” grounds the problem in “Many of the ideas in this lesson — and in this entire Agent Engineering chapter — come from her June 2023 blog post "LLM Powered Autonomous Agents" . A deeply respected pioneer, she laid out the profoundly infl…”. “Four Core Capabilities (from Lilian Weng's Architecture) · Click to Switch Demo” then moves it toward “🧭 Plan Breaks complex tasks into executable step sequences — think before acting 🔧 Tool Use Calls search, code execution, databases, and APIs — an extension of hands and feet 🧠 Memory Short-term context + lo…”. Together, they show that the lesson is not just a conclusion to remember, but a claim with conditions.

Carry the judgment into the next situation

The same logic applies to retrieval: define what counts as relevant, check whether recall covers the question, and then inspect whether ranking, chunking, or freshness pushed useful evidence out.

  • “The Classic Architecture · Where It All Comes From”: Many of the ideas in this lesson — and in this entire Agent Engineering chapter — come from her June 2023 blog post "LLM Powered Autonomous Agents" . A deeply respected pioneer, she laid out the profoundly infl…
  • “Four Core Capabilities (from Lilian Weng's Architecture) · Click to Switch Demo”: 🧭 Plan Breaks complex tasks into executable step sequences — think before acting 🔧 Tool Use Calls search, code execution, databases, and APIs — an extension of hands and feet 🧠 Memory Short-term context + lo…
  • “The closing point”: This one diagram has shaped the entire AI Agent industry — nearly every Agent framework today traces back to it. We highly recommend following her on Twitter/X

The final “The closing point” brings the discussion to “This one diagram has shaped the entire AI Agent industry — nearly every Agent framework today traces back to it. We highly recommend following her on Twitter/X”. The useful thing to carry forward is knowing which judgments must be revisited when input, scale, or risk changes.

Mark as learned Your reading progress updates automatically
← PreviousNext →

Keep reading

The next useful article in the thread.

ARTICLE DISCUSSION

Leave one useful thought here.

Keep the idea that clicked, the question that stayed open, or a small note for the next learner.

Discussing Agent: AI That Gets Things Done The Harness Around the Model
3discussionsArticle discussion · synced with the Circle
View in the learning circle
AM
Asha MorganContent editor
INSIGHTField note

I turned one judgment from this article into a small experiment I could run today. Knowing what to observe next is more useful than simply remembering the conclusion.

ARTICLE DISCUSSION7 helpful
LH
Lin HarperIndie developer
INSIGHTInsight

After reading this, I first looked for the conditions behind the idea instead of copying the method into a project. That order made the later trade-offs much clearer.

ARTICLE DISCUSSION5 helpful
KM
Kiki MooreProduct operations
QUESTIONQuestion

When this judgment reaches real work, which constraint should be added first? I am curious which step matters most between reading and the first practical attempt.

ARTICLE DISCUSSION4 helpful