Peter Steinberger's Loops or Graphs Tweet (2026)
Peter Steinberger's July 18 tweet did not coin graph engineering. It amplified an existing phrase. See the first four days of the July 2026 wave.
Course outline · Build AI Agents (4.21)
On July 18, 2026, Peter Steinberger posted one question to X: "Are we still talking loops or did we shift to graphs yet?" The post did not coin "graph engineering." The phrase appears in a 2024 post and again in a July 11, 2026 post. What Steinberger's question did was amplify it to a much larger audience. During the first four days after the post, the phrase spread through replies and X Articles. LangChain also published an official response.
If you searched the exact sentence, the short answer is that the post is real and the wording above is exact. This page tracks the July 2026 wave around it, not the origin of graph-shaped orchestration or the phrase itself.
What Did Peter Steinberger Actually Say?
The whole post is twelve words:
Are we still talking loops or did we shift to graphs yet?
Steinberger posted it at 00:34:54 UTC on July 18, 2026. The post contains neither the phrase "graph engineering" nor a claim about a new technique.
It reads as a joke about vocabulary churn. That is an interpretation of the wording, especially the word "yet," rather than a statement of Steinberger's intent. The source does not say what he expected the audience to do with it.
The Phrase Existed Before the Tweet
The clearest earlier use in the sources is from Itamar Friedman on February 29, 2024, in a reply to LangChain:
When building real world AI empowered systems, we see a shift from prompt engineering to flow (/graph) engineering.
One week before Steinberger's post, Mike (@michaelmasson55) posted this ladder:
prompt engineering
context engineering
harness engineering
loop engineering
next is: graph engineering
Those posts rule out an origin story centered on July 18. LangChain's own retrospective makes the narrower claim: the July discussion was kicked off by Steinberger's tweet. That describes the start of a viral wave around an existing phrase.
The same distinction matters for "loop engineering." Franziska Hinkelmann used and defined "Loop Engineering" on March 8, 2026. Addy Osmani later wrote the widely shared June article that organized the practice into a system and parts list. His article helped popularize and structure the term, but the term did not come from that article.
Who Posted It?
On his own About page, Steinberger says he made OpenClaw, joined OpenAI in February 2026, and previously bootstrapped PSPDFKit. Those facts establish who posted the question. They do not establish why it spread, what he intended, or which tools his audience had installed.
The evidence for the wave is in the dated responses themselves.
What Happened in the First Four Days After
Before July 18
Friedman's February 2024 post and Mike's July 11, 2026 post show that "graph engineering" was already in use. They belong at the start of this timeline because they separate earlier usage from Steinberger's later amplification.
Free AI Builder Newsletter
Weekly guides on AI tools & builder strategies.
Day 0, July 18
Steinberger posted at 00:34:54 UTC.
Shubham Saboo followed at 02:10:05 UTC with this quote-post:
Loops made Agent behavior programmable. Graphs make agent orgs programmable.
Hamel Husain published the X Article "Loop Engineering Is Dead. Enter Graph Engineering" at 05:09:36 UTC.
Ten minutes later, Yohei Nakajima quote-posted:
loops go oops
use graphs to laugh
Carlos E. Perez's "From Loop Engineering to Graph Engineering?" arrived at 10:00:04 UTC. It was a same-day long-form response, nearly five hours after Husain's article.
Day 1, July 19
Dale Everett published "Loops are just shitty graphs." at 22:25:05 UTC. His argument was that a loop is already a simple graph, which placed the July debate over old architectural ground.
Day 2, July 20
Harrison Chase, a LangChain cofounder from the team behind LangGraph, wrote at 15:00:06 UTC:
So i didn’t really know what graph engineering is, and i still don’t really… but it’s basically just langgraph?
The team-level attribution matters. LangChain's official About page says the company launched LangGraph; it does not credit Chase as its sole creator.
Day 3, July 21
Codila published a freely readable X Article at 16:02:12 UTC with "full 5-step course" in its title. The source supports one course-styled article. It does not support the earlier claim that multiple paid courses had appeared.
Day 4, July 22
Shann Holmberg published "graph engineering explained (marketing edition)" at 11:50:38 UTC.
Across the first four days, the record shows a fast sequence of posts around Steinberger's question. It does not show that the question originated the phrase.
Did Anything About Building Agents Change?
None of the cited July posts announced a new graph-orchestration capability. LangGraph, AutoGen GraphFlow, and Google ADK already supported graph-shaped orchestration. LangChain's retrospective says it had been building agents as graphs for years.
What changed in July was the phrase's reach. More people used "graph engineering" to discuss explicit nodes, routes, shared state, and control flow around agents. That wider circulation can make an existing design pattern easier to discuss without turning the phrase into a new capability.
The practical decision still depends on the work. Use a loop when one agent can pursue one objective and check its own progress against a clear verifier. Consider a graph when the job needs explicit branching, state shared between specialized steps, or controlled handoffs. The detailed mechanics belong in the graph engineering guide, not in an origin story about one tweet.
What Should You Do About the July Wave?
- Do not restructure a working system because the vocabulary moved. A new burst of discussion does not make a graph necessary.
- Learn the architecture before you need it. The graph engineering guide covers nodes, edges, and shared state.
- Use a concrete trigger. Move beyond one loop when the work needs branching or a handoff that must preserve state.
- Start with the smallest implementation. Graph Engineering with Claude Code shows how subagents can act as nodes without adding a new framework.
Related Content
- Graph Engineering: What Happens When Your Agent Outgrows the Loop - the pillar guide to nodes, edges, shared state, and frameworks.
- Graph Engineering vs Loop Engineering - the architectural differences behind the July vocabulary.
- Is Graph Engineering Just LangGraph? - a framework-by-framework answer to Harrison Chase's question.
- Graph or Loop: When to Use Which - a decision guide for choosing the smaller shape that fits.
- Loop Engineering Guide - how to build and verify a single agent loop.
Start Here
If you came for the July 18 post, the historical answer is now clear: Steinberger amplified "graph engineering"; he did not originate it.
Read the graph engineering guide for the mechanics, or join AI Builder Club to build one with other practitioners.
Frequently Asked Questions
What did Peter Steinberger actually say about loops and graphs?
On July 18, 2026, he posted one question to X: 'Are we still talking loops or did we shift to graphs yet?' That is the entire post. It did not include the phrase 'graph engineering,' a framework, or an announcement.
Did Peter Steinberger invent graph engineering?
No, and he did not claim to. Itamar Friedman used 'flow (/graph) engineering' in February 2024, and Mike (@michaelmasson55) posted a ladder ending in 'graph engineering' on July 11, 2026. Steinberger's July 18 question amplified the phrase and kicked off that weekend's wave.
Who is Peter Steinberger?
Steinberger says he made OpenClaw, joined OpenAI in February 2026, and previously bootstrapped PSPDFKit. Those facts come from his own About page. This article does not infer that his biography caused the post to spread.
Did anything about building agents change on July 18, 2026?
None of the cited posts announced a new graph-orchestration capability. LangGraph, AutoGen GraphFlow, and Google ADK already supported graph-shaped orchestration. The July discussion gave an existing phrase wider circulation.
Is graph engineering just LangGraph?
Harrison Chase, a LangChain cofounder from the team behind LangGraph, wrote: 'So i didn’t really know what graph engineering is, and i still don’t really… but it’s basically just langgraph?' LangGraph is one established implementation of the pattern, not proof that the phrase began in July 2026.
Sources & Verification
This is a sourcing piece about Peter Steinberger's July 2026 post and the wave it amplified. Every quoted post is tied to a dated primary source. Earlier uses from 2024 and July 11, 2026 are included to separate the phrase's history from the July wave. The timeline covers the first four days after Steinberger's July 18 post. Engagement is described qualitatively because X counts move after capture. See our editorial standards.
- Itamar Friedman (@itamar_mar), 2024 use of flow (/graph) engineering - Friedman's February 29, 2024 post said that real-world AI systems were shifting from prompt engineering to flow (/graph) engineering.
- Mike (@michaelmasson55), the engineering ladder ending in graph engineering - Mike posted the full prompt, context, harness, loop, graph ladder on July 11, 2026, one week before Steinberger's post.
- Peter Steinberger (@steipete), the July 18 question - Steinberger posted at 00:34:54 UTC on July 18, 2026: 'Are we still talking loops or did we shift to graphs yet?' This article tracks the wave around that question.
- Shubham Saboo (@Saboo_Shubham_) - Saboo quote-posted Steinberger at 02:10:05 UTC on July 18, framing graphs as making agent organizations programmable.
- Hamel Husain (@HamelHusain), Loop Engineering Is Dead. Enter Graph Engineering - Husain published the first long-form response in this timeline at 05:09:36 UTC on July 18, 2026.
- Yohei Nakajima (@yoheinakajima) - Nakajima posted the two-line 'loops go oops' and 'use graphs to laugh' quote-post at 05:19:20 UTC on July 18, 2026.
- Carlos E. Perez (@IntuitMachine), From Loop Engineering to Graph Engineering? - Perez published a same-day long-form response at 10:00:04 UTC on July 18, 2026. It links Steinberger's post and argues for an architectural shift.
- Dale Everett (@daleverett), Loops are just shitty graphs. - Everett published the counter-position that a loop is already a simple graph at 22:25:05 UTC on July 19, 2026.
- Harrison Chase (@hwchase17) - Chase, a LangChain cofounder from the team behind LangGraph, asked whether graph engineering was basically LangGraph at 15:00:06 UTC on July 20, 2026.
- Codila (@0xCodila), Graph Engineering: build 1000+ agent loops in one window, from one prompt (full 5-step course) - Codila's freely readable X Article, presented as a full 5-step course, appeared at 16:02:12 UTC on July 21, 2026.
- Shann Holmberg (@shannholmberg), graph engineering explained (marketing edition) - Holmberg's marketing-edition explainer appeared at 11:50:38 UTC on July 22, 2026, at the end of the four-day timeline.
- 3 Years of Graph Engineering with LangGraph (LangChain) - LangChain says the July discussion was kicked off by Steinberger's tweet while also stating that graph-shaped agent systems predated it.
- About LangChain - The official company history identifies Harrison Chase as a cofounder and says LangChain launched LangGraph.
- Franziska Hinkelmann (@fhinkel), Loop Engineering - March 8, 2026. Hinkelmann used and defined Loop Engineering three months before Addy Osmani's June article.
- Loop Engineering (Addy Osmani) - The widely shared June article that organized loop engineering into a concrete system and parts list.
- About Peter Steinberger - Steinberger's account of making OpenClaw, joining OpenAI in February 2026, and previously bootstrapping PSPDFKit.
- Graph Engineering Guide (AI Builder Club) - The pillar guide to nodes, edges, shared state, frameworks, and the choice between a graph and a loop.
Join AI Builder Club
$37/mo
Get the free newsletter
Weekly deep-dives on AI tools, automation workflows, and builder strategies. Join 5,000+ readers.
No spam. Unsubscribe anytime.