The AI Forward Deployed Engineer Playbook

The FDE Problem: Speed Without a Backlog

Forward deployed engineers do something rare: they sit with customers, understand specific workflows, and build solutions on-site. The problem is that building takes time and customers rarely have time. By the time a custom integration clears the engineering queue back at HQ, the customer has moved on or found a workaround.

AI tools, specifically Claude Code, MCP, and n8n, let FDEs compress weeks of work into days. Here is the playbook.

Rapid Prototyping with Claude Code

The first rule of FDE work: build the thing in the room. Claude Code lets you go from a described workflow to a working script in 30 minutes. Use this pattern at the start of every customer engagement.

Claude Code writes the script, tests it against a sample file, and handles edge cases like missing columns and network timeouts. You can demo this to the customer within the hour and gather real feedback before writing a single line of production code yourself.

Custom Integrations with MCP and n8n

When the customer needs something that persists beyond a one-off script, reach for MCP or n8n. MCP lets Claude interact directly with a customer's internal tools via a server you configure once. n8n handles multi-step workflows with a visual editor the customer can maintain themselves after you leave.

{
  "nodes": [
    {
      "name": "Schedule Trigger",
      "type": "n8n-nodes-base.scheduleTrigger",
      "parameters": {
        "rule": { "interval": [{ "field": "hours", "minutesInterval": 1 }] }
      }
    },
    {
      "name": "Read CSV from SFTP",
      "type": "n8n-nodes-base.sftp",
      "parameters": {
        "operation": "download",
        "path": "/exports/orders_today.csv"
      }
    },
    {
      "name": "Filter Delayed Orders",
      "type": "n8n-nodes-base.code",
      "parameters": {
        "code": "return items.filter(i => i.json.status === 'delayed')"
      }
    }
  ]
}

You can generate this n8n workflow JSON with Claude Code, import it into the customer's n8n instance, and show a live automated workflow in under 30 minutes. The customer owns it after you leave.

Building On-Site Tooling That Sticks

The best FDE work leaves customers with tools they actually use. That means simple interfaces, not dashboards that require a data team to maintain. For internal ops tooling, a small Streamlit app built with Claude Code often serves the team better than a complex BI integration.

A self-contained tool the ops team can run themselves is worth ten dashboards they need IT to update. FDEs who ship simple, maintainable tools build lasting customer relationships and reduce their own support burden.