Attachments are the quiet time sink in most workflows. Invoices arrive as PDFs, purchase orders as Excel, contracts as DOCX, receipts as photos — and somebody manually opens each one, keys the relevant fields into a system, and files the original. The 2026 answer to this isn’t OCR; it’s a document-aware agent that reads the attachment, understands what it is, and routes it accordingly.
Classifies the attachment. Invoice, purchase order, contract, receipt, shipping doc, lab report — the agent reads the file (PDF, DOCX, XLSX, image) and recognizes its type, not just its MIME type.
Extracts structured fields. Vendor, total, line items, due date for invoices. Parties, start date, auto-renewal for contracts. Sender, amount, category for receipts. Extraction uses modern document-AI models (Claude 4.7 Opus, GPT-5, Gemini 3 Pro — whatever fits the document type best), not regex over raw OCR. Vision-capable models read stamped, handwritten, and multi-column layouts that broke OCR pipelines as recently as 2024.
Routes to the right system. Invoices into your AP tool (Ramp, Brex, QuickBooks). POs into your ERP. Contracts into CLM. Receipts against the right expense report. Everything over MCP, everything logged.
Human in the loop where it matters. Amounts over a threshold, unfamiliar vendors, contracts mentioning indemnification — these pause for a human approver. Routine operations ship through untouched.
Files the original. The source PDF lands in Drive / Dropbox / your DMS, named and tagged so it’s findable six months later.
Two things changed. Document AI got good enough to handle messy real-world scans (not just pristine templates). And MCP made downstream routing easy — writing to a dozen systems no longer requires a dozen integrations.
Finance teams drowning in invoices. Legal ops handling contract intake. Ops teams receiving shipping docs. Anyone whose inbox is effectively a document queue someone triages by hand.
Pairs naturally with the Gmail Triage Agent if you also want inbox labeling and draft replies.
Pair with one of our solutions architects. Two weeks from kickoff to a deployed, evaluated, observable agent in your stack.