The answer is always a layer down.
Start at the fundamentals, change one thing, measure, repeat. Three systems below, with what broke, what I got wrong first, and what I'd do differently. Employers are anonymized; the engineering is not.
0 write tools given to the model
In acceptance testing, not production.
The model can't file anything
Enterprise support intake as one conversation — designed so the model is never the thing that creates the record.
capability boundary 94× faster — 30s to 318ms
Thirty seconds was the symptom
A notification service taking half a minute to respond, failing a third of the time. Both numbers had the same cause.
transaction scope 1.6M records moved, zero downtime
Moving 1.6 million records without downtime
A document portal moved between databases in production, with a rollback plan never more than a code revert away.
rollback designWorking on the same thing?
If your team is working through any of this, I'm happy to share what I've learned.