0
Pre-work
Capture requirements
0.1
AM
Gather requirements on a recorded call with the client. This recording becomes the source of truth for everything that follows.
1
Design
Remove every unknown before building
Replace any missing piece with a mockup (data, designs, folder structures, documents) so we can demonstrate the prototype in the shortest time possible. Iterate here where it's cheap, not later where it's expensive.
1.1
Dev
Feed requirements and call transcript to Claude. Build a working prototype in our N8N account (not the client's).
1.2
Dev
Use GWS CLI to mock up any missing data (PDFs, folders, sheets) on our Google Drive. Never build these manually.
1.3
Dev
Use Pencil to create professional mockups of outputs the client will see (emails, reports, dashboards).
1.4
Dev
Create a "How It Works" doc (.md) with a Mermaid diagram showing the full workflow.
1.5
AM
Review the How It Works diagram. Show the working prototype to the client. Collect feedback.
1.6
Dev
Iterate with Claude based on AM feedback until no questions or unknowns remain on the dev side.
1.7
AM
Get client approval. Client must sign off on: process flow, each workflow's logic, every output format, and edge case handling.
APPROVAL GATE -- No approval = cannot proceed to Development
2
Development
Build, test, and validate with client
We only enter this phase after 100% of the nuance has been answered. If unknowns surface, we go back to Design where iteration is faster.
2.1
Dev
Have Claude build a list of needed credentials.
2.2
Dev
Move or copy the prototype from our N8N to the client-owned N8N instance.
2.3
AM
Share the credential list and setup instructions with the client. Have them start setting things up.
2.4
Client
Begin setting up credentials on their end.
2.5
Dev
While client sets up credentials: create 10+ test scenarios using N8N evaluation nodes. Fix anything that breaks. Break into sub-workflows as needed.
2.6
Dev
Connect to client credentials and test in a sandbox (safe area, no live data).
2.7
Dev
Create a testing instructions doc (.md) for the client.
2.8
AM
Give the client ability to test in sandbox. Collect feedback.
Feedback routing: Big changes (fundamental logic, new requirements) -> back to Design. Small tweaks (wording, minor adjustments) -> stay in Development. Approval should come quickly since both sides have already tested.
APPROVAL GATE -- No approval = cannot proceed to Deploy
3
Deploy
Go live with confidence
Client has tested it, we've tested it, and we have final sign-off.
3.1
AM
Work with the client to pick a deployment date and time.
3.2
Dev
Have Claude build a dev plan (.md) and deployment checklist covering everything needed from develop to launch.
3.3
Dev
Update all docs (.md) and Mermaid diagrams to reflect the final state.
3.4
Dev
Confirm the error handler is set up and working.
3.5
AM + Dev
Deploy the solution in the live environment on the agreed-upon date and time.
How AI Makes This Process Viable
Claude scaffolds N8N workflows from the call transcript and requirements
Claude writes and maintains the How It Works document; one prompt updates the doc
Claude generates and updates Mermaid diagrams as the design evolves
Pencil + Claude creates output mockups (emails, PDFs, spreadsheets) in minutes
Claude generates test scenarios, including edge cases you might not think of
Claude creates deployment checklists and testing instructions, keeping docs current
All artifacts stay in sync because Claude produces docs, diagrams, and workflows together
Without AI, this process would be prohibitively expensive. With AI, the design phase takes 1-5 hours for an average workflow instead of days or weeks.