Capability change
- Before: Agents needed manually authored JSON schemas or code to specify how to parse and extract fields from documents.
- Enabling change: LlamaIndex released Retrieval Harness for filesystem‑style document tools and Conversational Extract that builds extraction schemas through dialogue.
- After: Agents can interactively create and run extraction pipelines by describing the desired fields in natural language, enabling on‑the‑fly schema generation.
Human consequence: Developers and analysts spend far less time authoring schema definitions and can prototype document‑processing flows directly in chat.
Bottleneck: Accuracy still depends on LLM understanding of the document structure and on the quality of the underlying parsing models.
Buildability: B3. Verification: source-inspected. Related shift: document-processing-shifts-from-hand-crafted-ocr-pipelines-to-llm.
The simultaneous release of Retrieval Harness and Conversational Extract shows a coordinated push to make document parsing conversational. This removes a long‑standing friction point—hand‑coding schemas—making it feasible to build flexible extraction agents quickly. Builders can now assemble pipelines that let users ask, “Give me the invoice total and date,” and the system will generate the needed schema on the fly. Watch for limits in schema complexity and for edge cases where the LLM misinterprets document layouts, which will drive future improvements in grounding and validation.
Published automatically by the New Runtime weekly knowledge desk from the week's evidence.
Supporting evidence
- Parse, Extract, Classify — now each with its own MCP (signal, 2026-09-26)
- Parse, Extract, Classify — now each with its own MCP (signal, 2026-09-26)
Provenance
Auto-created by the New Runtime weekly knowledge desk on 2026-09-28 from evidence reviewed in the week of 2026-09-21. It is published through the live overlay and joins the hand-edited catalog at the next release.