September 2026 Summaries

Vertesia’s September 2026 release expands the platform into a full environment for building, deploying, hosting, and operating AI-powered applications, including a preview App Builder that generates applications from natural-language descriptions. The update adds Grounded Extraction for document processing, providing source citations, confidence-based review flags, and auditability, alongside AI-assisted collaborative editing and configurable intake policies for different document types. New Content Views offer reusable, continuously updated search configurations, while attribute-based access control applies detailed policies based on user, content, and contextual attributes across reads, writes, collections, and machine callers. Vertesia also introduces a platform-wide event-driven architecture centered on an Event Bus that supports real-time subscriptions, webhooks, automatic indexing, and queryable audit trails, replacing legacy workflow rules and consolidating intake into a unified six-stage pipeline.

The post argues that traditional OCR-based intelligent document processing often struggles in production because real business documents commonly include handwriting, multi-page tables, changing layouts, and both native digital and scanned formats. It presents four cases where conventional page-by-page OCR and template-driven extraction can create errors or require manual intervention: interpreting handwritten notes and stamps, combining line items across page breaks, extracting fields from variable supplier layouts, and unnecessarily converting text-based PDFs into images for OCR. Vertesia positions its AI-powered approach, based on vision-enabled large language models and semantic document understanding, as an alternative that processes visual and native text signals together, recognizes document-wide structures, maps fields by meaning rather than fixed coordinates, and selects extraction methods based on file type. The post attributes these differences to modern AI-native architecture rather than legacy OCR systems retrofitted with semantic capabilities, and recommends evaluating IDP products using complex, unstructured documents rather than polished examples.