Same job. 6 AM.
Two agents.
An OpenAI dot and a Claude Code routine, compared
on one page, from a real side-by-side test.
The concierge
and the alarm clock.
If your day is inbox and Slack, a dot notices things for you. If you build things and want the same job done every night, a Claude Code routine already does it.
Who starts the work
A dot can start on its own: it looks through your connected apps (read-only) for ways to help. A Claude Code routine runs when you ask, on a schedule, on an API call or on a GitHub event.
Memory
A dot remembers you, but you can’t edit that memory line by line yet. A routine starts blank every run and reads the files you write, which you can open and fix.
Where it works
A dot has its own cloud computer and browser, and talks to you in ChatGPT, Slack or Teams. Claude Code works in your project and reaches other apps through connectors (MCP).
Price
Dots start with ChatGPT Pro at $100 a month, and aren’t in the UK, EU or Switzerland yet. Claude Code routines run on the $20 Claude Pro plan and use your normal limits.
Control
A dot has Custom Rules per action and always asks before passwords or money. A routine doesn’t stop to ask, so trim its connectors before it runs.
Give it the
smallest key.
A new dot inherits every app already connected to your ChatGPT account. A routine can use every connector you leave on it, writing included. Decide what it can reach before it starts.
Check what's already connected before you start. One project folder, not your whole drive. Read-only if it only needs to read. Money and passwords stay with you.
A note on the test
One job, two mornings, two agents. It’s a real side-by-side, not a benchmark. Prices and availability are as of early October 2026 and may change.
A companion to Build Alone Episode 06.