Building Business Automation Systems That Get Used
Automation fails when nobody owns the failure. A script without logs is a future outage. I build for recovery: retries, screenshots, webhooks, and a human inbox when the portal changes.
Operate it, or it dies
Anybot-style browser workflows look impressive in a demo and rot the first time a vendor moves a button. Production automation needs an owner, a log, and a path back to a person.
If the task is rare, do not automate it. If it is daily and dull, it belongs in a system, not a volunteer.
Playwright is not the product
The product is a reliable workflow: idempotent steps, screenshots on fail, alerts, and a queue so one timeout does not stall the rest of the day.
I ship automation from Tangier the same way I ship SaaS — observable, owned, and boring in the best sense.
Key takeaways
- Every automation needs an owner and logs.
- Automate daily dull work, not rare heroics.
- Design recovery before the happy path.
FAQ
API or browser automation?
Prefer an API when the vendor has one. Use Playwright when the work only exists in a UI. Never scrape a UI for data an API already returns.