Automation
This is what it looks like when I build for others.
Four examples from running projects. An agent here is a program that handles one work step on its own and hands back the result. Two of them run as demos behind a login, access on request.
Automation
From inquiry to quote
An informal request goes in, a finished quote comes out. Approval stays manual.
Inbox triage
Mail gets sorted and replies get drafted. Nothing is sent on its own.
The draft sits in your drafts folder, your edits change the rules.
Recurring work
What someone does by hand every week runs as an agent, with approval and an emergency stop that halts it at once.
Sending stops the moment a file exists on the server.
Working with agents in your team
One or two days on site, on your real backlog instead of examples.
Automation
- A call first: without the real workflow, any estimate is a guess.
- The human stays in the loop: the agent prepares, someone approves.
- Always an emergency stop: any workflow can be shut off in one move, with nobody touching code.
- Nothing on spec: what gets built is the step that eats time today.
That is the state of my own projects. What carries over to yours comes out of a conversation.
Questions that come up often
How does a project like this run?
A call first: what happens today, and what of it repeats. Then I build the smallest piece that already takes work off your plate. You approve it.
What happens to my data?
It stays where it is. What an agent may see is set upfront. Every step can be read back.
What does it cost?
That is not on this page, because it depends on scope. The honest first step is a call, not a price tag.
Does it then run without me?
Only as far as you want it to. Approval and the emergency stop stay with you, even when the workflow starts on its own every day.