AI Automation en San Francisco
How we work in San Francisco
A model mixes both forms inside one paragraph
If you generate replies, summaries or email drafts, the spelling comes from whatever the model has read, not from your style guide. You pin it in the system instruction and you still check afterward. The instruction shrinks the problem; it does not close it.
The check is a word list, not a judgment
Text can be run against a list of ruled-out forms before it goes out. It is twenty lines of code that finish in a second and catch what nobody was going to reread. It holds no opinion on style: it flags that a word is there that should not be.
Bulk replacement breaks correct sentences
In British English “programme” is the general word and “program” the computing one, and that pair has no mechanical rule. Search and replace in either direction produces errors no spell checker flags, because both words exist. It gets done with human review.
Automated emails are what everyone forgets
Confirmations, reminders and workflow notices get written once, copied from an old template and left there for years. When we audit the spelling of a whole site, a good share of the stray occurrences are sitting in them. They belong in the same string catalog as the screens.
Answering your questions
What people ask us about AI Automation in San Francisco.
If your question is not here, tell us about the project and we will answer with real context.
Ask us directlyReal work
Projects we have built
Full service · All the detail
AI Automation: stack, process, use cases and FAQs