Week 44 / 2026.
A short record of changes between issues. Full answers and charts live on the question pages.
- Data window
- Oct 26, 2026 – Nov 1, 2026
- Last updated
- Nov 13, 2026
- Author
- Romuald Członkowski
Commentary
Between 26 October and 1 November, n8n-mcp produced 73,474 workflows, 10,496 a day on average, across 1,833,016 tool calls with a 0.10% error rate. That is 12% fewer workflows a day than in the previous issue and 1% fewer calls, against 159,316 active installs, down 3%; 8% of them create workflows. The busiest day was 29 October with 13,312 workflows, the quietest 1 November with 5,348. Week-one retention stands at 33% of the 6,717 installs that started in the week of 19 October, 14% below the week before. The data gives no reason for any of these drops, so I am not guessing at one.
Of the calls, 55% are reads, 12% writes, 7% documentation and validation, and 4% tests. Fetching executions alone accounts for 34% of all calls. The window holds 144,829 edits: all of them went through the API, 54% produced a verifiable improvement, at an average of 2.61 operations and 1.42 changed properties with a median of 872 ms. What usually changes is a node that already stands there: updateNode 45,501 times, patchNodeField 42,179, and among the known edit purposes a configuration change leads (53,565) ahead of adding functionality (18,791). Every new workflow comes with 7.7 updates, so when you price an implementation, most of the work falls after go-live rather than before it.
The window added 27,871 workflows to the corpus, which now holds 1,276,985, with a median of 7 nodes. Code appears in 74% of them, HTTP Request in 62%, Data Table in 13.0%, an LLM model node in 6% and an AI Agent in 3%, while 14% call a sub-workflow. Since March I have been seeing the same direction: the Data Table share has climbed from 3.3% to 13.0% in the November data, and the model-node share has fallen from 17% to 5%. Data table is also the week's most searched term, with 183 installs behind it, ahead of webhook (127). Among model nodes, OpenAI holds 33%, Anthropic 20%, Google Gemini 18% and OpenRouter 14%. A webhook is the trigger in 45% of workflows, a schedule in 22%.
Validation turned away 0.7% of workflows, and again it is the smallest ones that trip over it: 0.99% at 2–3 nodes, against 0.12% at 31–100. Day by day the rate stayed between 0.46% and 0.89%. 7% of connection tests end in an error, and 37% of those come down to an address or the network. This issue's pitfall is marked critical: if a Wait node sits in a sub-workflow called through Execute Workflow, data can be lost on resume, because the sub-workflow does not always return it to the parent. Build that flow on webhooks instead of a Wait node.
Changes versus the previous issue
| figure | value | change |
|---|---|---|
| Workflows created | 73,474 | — |
| Workflows per day | 10,496 | -12% |
| Tool calls | 1,833,016 | -1% |
| Calls ending in error | 0.10% | -17% |
| Failed validation | 0.7% | -1% |
| Active installs | 159,316 | -3% |
| Week-1 retention | 33.1% | -14% |
| Workflows added to corpus (week) | 27,871 | -5% |
This week's answers
- Which LLM powers n8n AI agents: OpenAI, Anthropic, Gemini or Ollama?
In n8n workflows built last week, OpenAI holds 33% of model nodes, Anthropic 20%, Google Gemini 18% and OpenRouter 14%. In Mar 2026 OpenAI had 48% and Anthropic 12%.
- Which n8n nodes do AI agents use most, and which trigger nodes start their workflows?
The most used node in AI-built n8n workflows is Code, present in 74% of last week's workflows, followed by HTTP Request (62%) and Webhook (50%). The median workflow has 7 nodes.
- Which n8n node is growing fastest?
The fastest-growing n8n node is stopAndError: its share of AI-built workflows rose from 0.5% in Jul 2026 to 3.1% in Oct 2026, a 6.7× increase. Extract from File is second (3.3×).
- Do AI-built n8n workflows handle errors?
Rarely. 2.2% of AI-built n8n workflows from last week contain an Error Trigger node, and 2.4% of those built in Oct 2026. Since March the share has never exceeded 2.5%.
- Do AI agents build n8n workflows from nodes or from code?
From code. The Code node appears in 74% of AI-built n8n workflows from last week and HTTP Request in 62%. In Oct 2026 the Code share was 76%. Agents would rather write JavaScript and call an API than look for a ready-made integration.
- How large are AI-built n8n workflows, and which fail validation?
52% of workflows created last week have at least 11 nodes and 19% have 31 or more. The larger the workflow, the less often it fails validation: 1.0% for 2–3 nodes versus 0.12% for 31–100.
- Which n8n nodes do AI agents search for?
Last week the most searched n8n node was data table (183 installs), ahead of webhook (127) and telegram (42). The ranking counts unique installs, not queries.
- How do AI agents (Claude Code, Cursor) build n8n workflows through MCP: build or maintain?
Maintain. Of 1,833,016 tool calls last week, 55% read workflows, executions and lists, and 12% write. Fetching executions alone is 34% of calls. In the same week 73,474 workflows were created, 10,496 a day.
Most searched nodes
data table (183) · Webhook (127) · Telegram (42) · Slack (37) · whatsapp (36) · Gmail (34) · anthropic (32) · mcp client (32) · google sheets (31) · schedule trigger (31)
Execute Workflow + Wait nodes cause data loss
When using Execute Workflow node with Wait nodes in the sub-workflow, data can be lost on resume. The sub-workflow execution may not properly return data to the parent workflow after the wait period. Avoid using Wait nodes in sub-workflows called via Execute Workflow — use webhook-based patterns instead.