<- all tokdocs
2026.08.27@agenticamit

Graph Engineering Is the AI World's Newest Must-Learn Term, and the Naming Treadmill Is the Real Story

AI
Watch on TikTok

View on TikTok ->

In 24 seconds, Amit delivers a deadpan PSA announcing that everyone in AI now has to learn "graph engineering," the successor to loop engineering, harness engineering, and memory engineering. The video is satire about how fast the AI industry mints new engineering disciplines, but the props he flashes on screen are built on real material, including Anthropic's actual knowledge graph cookbook and the long lineage of graph theory in computer science. The title he gave the video, "YAET: yet another engineering term," removes any doubt about the intent.

A PSA Delivered With a Straight Face

The video opens on a standard talking-head shot with "Graph Engineering" as the title card and the caption "Just a quick PSA to" appearing below. Amit keeps the tone of someone sharing urgent career advice. The script is one long run-on chain: graph engineering extends loop engineering, which extends harness engineering, which extends memory engineering, "and you get the point." That escalating chain is the joke. Each term sounds plausible on its own, and stacking four of them in one breath exposes how the naming cycle works.

The Mock Paper Sells the Bit

The strongest visual gag lands around the six-second mark. Amit holds up a document styled like a research paper, headed "Agentic Software Engineering Practice 2026." The title reads "Graph Engineering: The Karpathy Loop, Improved 1000x by Itself. The Anthropic Playbook." The fine print says it is "Based on Anthropic Knowledge Graph Construction Cookbook and Andrej Karpathy courses and presentations available in the public domain," followed by a disclaimer that it was "Independently compiled, July 2026 - not affiliated with Andrej Karpathy and Anthropic - and not endorsed." Below the title sits a hand-drawn node diagram with an "AGENT" box in the center and arrows feeding in from boxes labeled Tools, Agency/Group, Extra Person Entity, and Extra Property/Attribute. The diagram parodies the crude whiteboard graphics that accompany every new AI framework announcement.

The Infographic Almost Makes the Case For Real

Midway through, a cleaner slide appears titled "The Big Picture" with the tagline "Three layers. Three jobs. One system." It lays out a three-stage pipeline: Harness (build the environment around the model), Loop (design the repeated work-and-feedback cycle), and Graph (make the workflow topology explicit and controllable). A second row maps the same flow as Environment, then Feedback, then Flow. Stripped of the satire, this is a reasonable description of how agentic systems have evolved: first tooling and sandboxes around the model, then iterative feedback loops, then explicit orchestration of multi-step workflows. The video works because the frameworks it mocks are close enough to real practice to be believable.

The Punchline: Graph Theory Always Comes Back

Amit closes by noting that "just as history predicts, graph theory comes right back into the fray to make things just that much better." That line carries the sharpest observation in the video. Graph theory has cycled through computer science repeatedly: social network analysis, PageRank, graph databases, graph neural networks, and now knowledge graphs for agent memory. The satire targets the packaging, not the substance. Anthropic really does publish a knowledge graph construction guide in its cookbook, covering entity extraction, resolution, assembly, and multi-hop querying with Claude. The underlying techniques are useful. The ritual of rebranding each one as a new engineering discipline you "have to learn" is what gets skewered here.

Key Takeaways

  • The video satirizes term inflation in AI, chaining four invented disciplines (memory, harness, loop, graph engineering) into one lineage to show how arbitrary the naming cycle is.
  • The mock paper prop mimics real hype patterns: a grand title, name-dropped authorities (Karpathy, Anthropic), a hand-drawn diagram, and a quiet non-affiliation disclaimer.
  • The "Harness, Loop, Graph" infographic is close to a genuine description of agentic system evolution, which is why the parody lands.
  • The real material behind the joke exists: Anthropic publishes a knowledge graph construction guide, and graph theory keeps returning as a foundation for new AI tooling.
  • Treat new "X engineering" terms as marketing labels first. Evaluate the underlying technique, since the label will change before the technique does.

Resources

Published August 27, 2026. Writeup generated from a favorited TikTok.