#agents

AI agents and automations

The work that arrives every day in the same shape: supplier invoices, order questions, support tickets, shipment updates, the numbers for Monday. An agent reads each item, does the part that is only tedious, and stops for a person at the line you set: an amount that differs, a refund, a customer who is angry. Built in n8n or Make with Claude or OpenAI behind it, running inside your inbox, your sheet and your Slack.

Will it break in six months? Every agent logs each run, retries on failure and posts errors to your Slack instead of failing quietly. You get the documentation and the credentials, and a monthly retainer if you would rather we watch it.

Typical deliverables

  • Invoice monitor with a hold rule
  • Inbox agent for order questions
  • Support agent that escalates
  • Daily ops summary in Slack
  • Shipment and delivery monitor
  • Prompts and guardrails, documented

Tools

  • n8n
  • Make
  • Zapier
  • Claude
  • OpenAI

Point of view

How we decide what to automate

Where a wrong number could reach a customer, the AI checks and a person decides. Where the work is only tedious, we automate it fully. Some things stay manual.

Automate fully

  • File incoming invoices in the right folder and pre-fill the expense sheet
  • Answer "where is my order?" emails with the tracking link from the shop
  • Post the day's open tickets to Slack at 9:00, oldest first

AI checks, human decides

  • Invoice total differs from the order by more than 2 %held for finance
  • A support reply that mentions a refund or a complaintdrafted, a person sends it
  • A supplier asks to change delivery termsreply drafted, you confirm it

Leave it manual

  • Complaints from your ten biggest customersa named person answers, always
  • Approving any payment or refund, whatever the amount
  • Anything that happens twice a yearnot worth the build