AI coding agent workflow

An AI coding agent workflow is a flowchart of one turn: the user prompt is assembled into context, the model either answers or calls a tool, and unsafe tool calls require a human to give a green light to go ahead. A denied tool, a failed tool, a full context window, and a failing test all return to the same place — another model call — which is why the picture is a set of loops around the model call hub, rather than a pipeline.

People draw this one to explain what an agent is actually doing while it sits on a prompt, to review which tools are allowed to run unattended, or to show where a turn spends its time. There's sufficient complexity here to make this difficult to explain in text, so the picture really is worth a thousand words.

Open in editor

Mermaid source

---
title: AI coding agent workflow
---
flowchart LR
  prompt[/User prompt/] --> gather[Assemble context]
  gather --> model[Call the model]
  model --> next{Tool or answer?}
  next -->|Tool| safe{Safe to run?}
  safe -->|No| human[Ask the human]
  human --> ok{Green light?}
  ok -->|Denied| model
  ok -->|Yes| which{Which tool?}
  safe -->|Yes| which
  which -->|Read| read[[Read files]]
  which -->|Edit| edit[[Edit code]]
  which -->|Shell| shell[[Run commands]]
  read --> outcome{Did it work?}
  edit --> outcome
  shell --> outcome
  outcome -->|No| log[Capture the error]
  log --> model
  outcome -->|Yes| add[Add result to context]
  add --> room{Context full?}
  room -->|Yes| squash[Summarise the history]
  squash --> model
  room -->|No| model
  next -->|Answer| draft[Draft the answer]
  draft --> tests{Tests pass?}
  tests -->|No| model
  tests -->|Yes| reply[/Reply to the user/]
  reply --> more{Another ask?}
  more -->|Yes| gather
  more -->|No| done([Turn complete])

Stock Mermaid vs Line9 on this AI coding agent workflow

Run the same source through the stock Mermaid engine and it often will not look as good. In some cases, Mermaid is able to deliver a usable graph, but not always. On this one:

The same AI coding agent workflow through stock Mermaid — long return arcs into Call the model, with Summarise the history stranded on the bottom-right — Stock Mermaid · same source View full size ↗

The most common fault of the stock Mermaid renderer is elongation: stretching the flowchart out into a long, narrow shape. In this example the diagram is stretched horizontally, forcing the multiple arrows that loop back to ‘Call the model’ to travel all the way from the right-hand side of the canvas. The dagre renderer's layout also produces several crossing lines which are not necessary. Line9 achieves a much more usable aspect ratio of 2.2:1, has edges that are far simpler to follow and has no edge crossings.

For a fuller product comparison — layout, export, CLI, and pricing — see Line9 vs mermaid.live.

Render your own

Paste any Mermaid flowchart into the free online editor — no account needed. Prefer the terminal? Install the line9 CLI (free for personal use).

More scenarios on the Mermaid examples hub.

Mermaid source

Stock Mermaid vs Line9 on this AI coding agent workflow

Render your own

Related diagrams