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
| Date | Source |
|---|---|
| 2026-07-26 | Gao Dalie — FORGET Loop Engineering. Graph Engineering is about THIS |
| 2026-07-29 | AI Labs — Anthropic Just Fixed Graph Engineering’s Greatest Flaw |
| 2026-07-30 | Caleb Writes Code — Graph Engineering explained in 8min |
| 2026-08-06 | Harshit Tyagi — After Loops We have Graph Engineering |
| 2026-08-09 | AI With Surya — Your Graph Engineering Is Probably a Loop |
| 2026-08-25 | Discover AI — AI Beyond Agent LOOPS: The Rise of Graph Engineering? |
| 2026-08-26 | AI 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.
Related
- Loop Engineering · Gao Dalie · Discover AI · ai-agentic-systems