Graph Engineering

Successor framing to Loop Engineering: model agent execution as a graph of
states and transitions rather than a single iterating loop.

The core claim

A loop is a graph with one node. Graph engineering argues that once agents branch, retry, fan out
to sub-agents or converge results, the loop abstraction hides the structure that actually
determines behaviour — so design the graph explicitly.

Timeline

DateSource
2026-07-26Gao Dalie — FORGET Loop Engineering. Graph Engineering is about THIS
2026-07-29AI Labs — Anthropic Just Fixed Graph Engineering’s Greatest Flaw
2026-07-30Caleb Writes Code — Graph Engineering explained in 8min
2026-08-06Harshit Tyagi — After Loops We have Graph Engineering
2026-08-09AI With Surya — Your Graph Engineering Is Probably a Loop
2026-08-25Discover AI — AI Beyond Agent LOOPS: The Rise of Graph Engineering?
2026-08-26AI Paper Slop — Graph Engineering in the Era of LLM Agents (paper)

Status of the debate

Not settled. AI With Surya’s “Your Graph Engineering Is Probably a Loop” (2026-08-09) argues much
self-described graph engineering is loop engineering relabelled, and Discover AI frames it as a
question rather than a conclusion. Treat the two terms as an active argument about agent
architecture, not a succession.

Create and read this note together with Loop Engineering — neither is
comprehensible alone.