Are there similar scripting-over-LLM approaches that could apply to tax work involving sensitive documents like PDFs?

One approach discussed is locally redacting sensitive information (e.g., Social Security numbers) out of PDFs before doing anything with them, so that once redacted there's more confidence in sending the remaining data to an LLM. More broadly, the same principle from bookkeeping may apply to tax: it's worth asking how much of a given tax task actually requires an LLM versus how much could be handled with a script instead, though this wasn't fully resolved.