November 21, 2026, (Inside AI) — Amazon Web Services has released a new skill for AI agents within Amazon Textract Core. The skill allows agents to extract text, tables, forms, and other data from documents without requiring teams to run servers or deploy infrastructure. The company announced the update in a blog post, positioning it as a way to simplify document processing for agent-based workflows.
The skill guides an agent to the correct Textract API for each document type. It also includes a tested parser that converts raw output into reading-order text and structured fields. These fields can include answers to questions, receipt and invoice details, and identity document fields. Previously, teams connecting an agent to Textract had to build and maintain their own integration. The new skill removes that burden.
The release matters because document processing remains a stubborn bottleneck for enterprise AI adoption. Many organizations have valuable data locked in PDFs, scanned forms, and images. Extracting that data reliably has required custom code and ongoing maintenance. By packaging the integration as a reusable skill, AWS reduces the engineering overhead for teams building agentic workflows. The move also signals a broader shift toward modular, composable AI capabilities that agents can discover and load on demand.
How the Skill Handles Confidence and Review
The skill processes multi-page PDFs and returns every page in reading order. It uses confidence scores to accept high-confidence results and flags the rest. Flagged items can be routed to a human review process. For example, a user can ask an agent to turn a patient intake form into structured data and flag unclear fields for staff review. The agent responds to natural language requests, returning the answers the user needs.
The skill is available at no additional charge. Standard Amazon Textract pricing applies. It is available in all commercial AWS Regions where Textract is available, including US East (Ohio), US East (N. Virginia), US West (N. California), US West (Oregon), Asia Pacific (Mumbai), Asia Pacific (Seoul), Asia Pacific (Singapore), Asia Pacific (Sydney), Asia Pacific (Tokyo), Canada (Central), Europe (Frankfurt), Europe (Ireland), Europe (London), Europe (Paris), Europe (Spain), and Europe (Zurich).
For teams using the AWS MCP server with the aws-core plugin, agents can load the skill automatically. The skill is also published in the AWS Agent Toolkit, so agents including Kiro, Claude Code, and Cursor can use it. AWS directs users to the AWS Agent Toolkit on GitHub for more details.
The introduction of a skill for Textract fits a pattern of cloud providers abstracting away infrastructure complexity. By offering a prebuilt, tested parser and API routing logic, AWS lowers the barrier for developers who want to add document intelligence to their agents. The confidence-based flagging mechanism also addresses a key concern in enterprise AI: reliability. Rather than forcing all results through a single automated path, the skill supports a hybrid workflow where humans review uncertain outputs.
This approach could accelerate adoption in regulated industries such as healthcare and finance, where accuracy and auditability are critical. The patient intake form example illustrates the potential. Staff can focus on flagged fields rather than manually rekeying entire documents. The skill's availability across all commercial Textract regions means global teams can deploy it without regional workarounds.
The AWS Agent Toolkit, where the skill is published, appears to be a growing collection of resources for building agent-based applications. By supporting agents like Kiro, Claude Code, and Cursor, AWS is making its services accessible through popular development environments. This interoperability could encourage more developers to experiment with Textract in agentic contexts.
Read: AWS Step Functions workflows now available as AI agent tools
The no-additional-charge model for the skill, combined with standard Textract pricing, keeps costs predictable. Teams pay only for the underlying Textract usage. The skill itself adds no premium. This pricing strategy may help AWS compete with other document AI providers that charge separate fees for integration tools or workflow features.
Looking ahead, the success of this skill will depend on how well it handles diverse document types and edge cases. The tested parser and confidence scoring are promising, but real-world documents vary widely. AWS will likely gather feedback and iterate. The company has not announced specific performance benchmarks or customer case studies yet. For now, the skill is available for immediate use, and AWS encourages developers to explore it through the Agent Toolkit.
Shamil has a low tolerance for AI that only works in controlled demos. He covers agentic systems and generative AI with the skepticism of someone who has seen too many “revolutionary” tools quietly disappear six months after launch.