Amazon Just Made Its Document-Reading AI Swappable in Seconds

Amazon Textract is Amazon's AI service for reading documents — it pulls text, handwriting and table data out of scanned invoices, mortgage applications, insurance claims and ID cards so people don't have to type it all in by hand. Companies can now train "adapters" (custom settings that teach the AI your particular form layout) so it reads your paperwork more accurately. This week Amazon published a how-to guide for managing those adapters once you move from testing into real, everyday production use.

The most useful part is a small trick with an outsized payoff. Instead of hard-coding which adapter an app should use, you store that reference in a central settings locker (AWS Systems Manager Parameter Store). When you swap in an improved adapter, the live system picks it up instantly — no downtime, no rebuilding the app. Amazon says a change that used to take hours or days of coordination now takes seconds. The guide also includes ready-made infrastructure templates, security settings for regulated industries, and a way to auto-sort documents so the right adapter handles the right form.

The honest catch: moving a trained adapter from a test account into a production account still requires filing an AWS support ticket, and only the model's learned weights transfer — your example documents and query definitions don't come along. There's also a hard limit worth knowing: Textract handles one adapter per page per API call, so if you process multiple versions of the same form, you must build routing first. It reads JPEG, PNG, PDF and TIFF files, up to 3,000 pages in the batch mode, but not XFA-based PDFs.

Bottom line: this isn't a shiny new product you'll sign up for. It's plumbing for the teams building document automation. You'll feel it indirectly — faster insurance claims, quicker loan approvals, fewer transcription errors, and shorter waits when a company changes how it processes your paperwork.