Question 1

What is the difference between prompt, loop and graph engineering?

Accepted Answer

Prompt Engineering designs a single instruction for a single response. Loop Engineering designs a repeatable execution cycle — plan, execute, validate, retry — with explicit stop conditions. Graph Engineering designs the whole system: multiple specialised nodes, routing between them, shared state, validators and human approvals.

Question 2

When should I use a prompt instead of a loop?

Accepted Answer

Use a prompt when the task is well defined, finishes in one pass, and you can judge the result yourself. Drafting, summarising, rewriting and one-off analysis rarely need more structure than a well-built prompt.

Question 3

When is a loop better than a prompt?

Accepted Answer

Use a loop when quality matters more than speed and the first output is usually not good enough. A loop adds acceptance criteria, self-checking and bounded retries so the work improves across iterations instead of needing manual re-prompting.

Question 4

When do I need a graph instead of a loop?

Accepted Answer

Use a graph when the work splits into different kinds of jobs that need different handling — research, drafting, review, approval — or when routing depends on intermediate results. A graph makes those paths explicit rather than hiding them inside one long loop.

Question 5

Can a prompt be converted into a loop or a graph?

Accepted Answer

Yes. Loop Engineering AI converts between the three forms in either direction, so you can start with a prompt, promote it to a structured loop, and expand it into a multi-node graph as the work gets more demanding.

Question 6

Do I need to learn all three disciplines?

Accepted Answer

They build on each other. Prompts are the instructions used inside loop and graph nodes, so prompt skill carries forward. The in-app Academy has a separate track for each discipline in that order.

Attribute	Prompt Engineering	Loop Engineering	Graph Engineering
Unit of design	A single instruction	An execution cycle	A system of nodes and edges
Control flow	Straight through, one pass	Repeat until criteria met or budget spent	Conditional routing, branches and parallel paths
Quality control	Human review after the fact	Built-in validation and bounded retries	Validators, quality gates and human approvals
State	Whatever is in the prompt	Carried across iterations of one cycle	Typed shared state every node reads and writes
Failure handling	Re-prompt manually	Retry strategy with stop conditions	Fallback paths, recovery nodes and safe termination
Best for	Fast, well-defined tasks	Quality-critical repeatable work	Complex multi-role workflows
Main limitation	No self-correction	Single kind of job per loop	Highest up-front design cost

Prompts Give Instructions. Loops Control Execution. Graphs Design Systems.

Three disciplines, three jobs

Prompt Engineering

Loop Engineering

Graph Engineering

How they differ, attribute by attribute

Prompt Engineering

Loop Engineering

Graph Engineering

Design all three in one place

Prompt, loop and graph engineering FAQ

Related pages

Pick the right level of structure