Recording intelligenceAI-generated brief · check the source for context

Graph Engineering

26:28 recording · AUTO · 1 speaker

Watch the original

Executive Summary AI
  • Greg Eisenberg says graph engineering, unlike most viral AI terms, is genuinely useful because it gives a better way to think about how AI work actually gets done, and promises listeners they will be able to turn one workflow they already run into a map of steps, checks, handoffs, loops and human approvals.
  • He defines it against its neighbours: prompt engineering is asking a better question, context engineering is giving better information, and graph engineering is designing the work around the AI so a startup-idea question is not decided, researched, written and graded by one model in one pass.
  • He separates the two things people mean by graph, a knowledge graph that helps AI reason over relationships (like Microsoft Graph RAG) and an agent graph that describes how work moves, and says this episode is about agent graphs because founders, creators and small teams can use them today.
  • Using the question of whether to launch an AI bookkeeping product for Shopify merchants, he walks a diamond shape: a planner splits the question, three researchers on customers, competitors and distribution run in parallel, a skeptic attacks stale or unsupported claims, a merge writes the recommendation, and a human gate decides what to test.
  • He warns against starting with Langraph or Autogen: level one is running the lanes manually on an Excalibur or TL draw board, level two is a repo where each step writes plan.md, customer.md, review.md and recommendation.md, tools come after the workflow, and the goal is the smallest graph that improves the quality of work.

Brief overview

Graph engineering means designing AI work as jobs, arrows, checks and human approvals instead of one chat.

  1. Separate the worker from the checkerAI research fails when the same model that writes the answer also grades it; in a good graph, checking is its own job.
  2. Draw the graph before you automate itRun the lanes manually three times first; if the manual version is not better, automation only produces mediocre work faster.
  3. Build the smallest graph that improves qualityMore agents can mean more noise and five workers repeating the same wrong idea; put human gates where mistakes are expensive.

Questions this recording answers

10 questions, each answered where it is said
How is graph engineering different from prompt engineering and context engineering?

Prompt engineering is how you ask the AI a better question, and context engineering is how you give it better information. Graph engineering is how you design the work around the AI, so the whole task stops living inside one messy giant AI chat and becomes a workflow you can manage.

What does a graph mean in graph engineering?

A graph is basically jobs connected by arrows. Each job is a step in the workflow, the arrows show what happens next, and the shared notes moving through the workflow are the state, meaning what the system knows so far. That is how real work like customer support already gets done.

What is the difference between a knowledge graph and an agent graph?

A knowledge graph helps AI reason over relationships, like which customer works at which company or which team owns a feature, where normal RAG struggles; Microsoft GraphRAG is an example. An agent graph is about how work moves: planner, parallel researchers, skeptic, synthesizer, then human approval. The best systems use both.

When should you use graph engineering, and when is it unnecessary?

Use a graph when work has multiple steps, some steps can run at the same time, and the output needs checking before it matters, such as deep research, go-to-market plans, support triage, code review or sales prep. Brainstorming ten project names or summarizing a short email probably does not need one.

How would a graph evaluate an AI bookkeeping product for Shopify merchants?

A planner splits the question into customer pain, competitors, go-to-market wedge, pricing and risk. Parallel researchers study merchants' bookkeeping habits, existing tools and freelancers, and where merchants hang out. A skeptic attacks weak claims, a merge step writes a pursue, pause or kill recommendation, and a human decides next steps.

Why does the graph need a separate skeptic or checker step?

A lot of AI research fails because the same model that writes the answer also grades it, like writing your own performance review. The skeptic asks which claims are supported, which evidence is stale, which competitor is ignored, and where pain is confused with willingness to pay. In a good graph, checking is its own job.

Do you need frameworks like LangGraph or AutoGen for your first graph?

No. Start by running the graph manually with separate lanes for each job, and draw it first on a blank Excalidraw or tldraw board. It is slower but easier to understand, and if the manual version does not produce better work, automating it just produces mediocre work faster. Automate after three successful manual runs.

What do the beginner, intermediate and advanced levels of graph engineering look like?

Beginner is a manual run with separate lanes. Intermediate uses Claude Code, Codex or a repo where each step writes files like plan.md, customer.md, review.md and recommendation.md, leaving a paper trail. Advanced orchestrates with LangGraph, AutoGen GraphFlow, n8n, Make.com or your own scripts.

What does a customer support graph look like?

It classifies the issue as billing, product confusion, bug or cancellation risk, checks account context like customer value and history, searches docs or internal policies such as a wiki or Notion, drafts a reply, has a checker review accuracy, tone and risk, and sends refunds, account changes, angry customers or legal risk to a human.

Do more agents in a graph produce better output?

No. More agents can mean more noise, five AI workers confidently repeating the same wrong idea, or more time coordinating than thinking. The goal is the smallest graph that improves quality: remove fake waiting, separate workers from checkers, put human approval where mistakes are expensive, stop when good enough, and keep useful state.

Key Quote

“But graph engineering is how you design the work around the AI so the whole thing stops living inside one messy giant AI chat.”

— Greg Eisenberg, Graph Engineering, at 1:40▶ watch at 1:40
Key Quote

“A lot of AI research fails because the same model that writes the answer also grades the answer.”

— Greg Eisenberg, Graph Engineering, at 12:01▶ watch at 12:01
Key Quote

“That is like asking someone to write their own performance review and then being shocked when they describe themselves as a visionary.”

— Greg Eisenberg, Graph Engineering, at 12:09▶ watch at 12:09
Key Quote

“And if the manual version doesn't produce way better work, automating it honestly will just produce mediocre work way faster.”

— Greg Eisenberg, Graph Engineering, at 14:19▶ watch at 14:19
Key Quote

“The first rep is to draw the graph before you automate the graph.”

— Greg Eisenberg, Graph Engineering, at 14:28▶ watch at 14:28
Key Quote

“The tool should come after the workflow. If you automate a workflow, you do not understand, you get a mess.”

— Greg Eisenberg, Graph Engineering, at 16:47▶ watch at 16:47

Cite this

“Graph Engineering.” https://www.youtube.com/watch?v=JWhICz1QR8M. Transcript summary by WhipScribe, https://whipscribe.com/transcript-pages/graph-engineering.html.

Report this page
What is wrong with this page?