GoKawiilNew Paper Office Python packages improve AI agents' Office file editing accuracy
A developer has released Paper Office, a suite of Python packages (paper-docx, paper-pptx, paper-xlsx) that let AI agents edit Word, PowerPoint and Excel files, building on the older python-docx, python-pptx and OpenPyxl libraries. In tests across five models and 61 tasks, Paper packages with guidance succeeded in 92.5% of trials, compared with 80.7% for the legacy libraries alone and 69.5% using Anthropic's Office skills, while agents resorted to editing raw file internals far less often (1.6% versus 78.7% and 50.5%).
GoKawiil's interpretation of the reporting above, not reported fact.
Office files are ZIP-based XML structures where small edits can require multiple consistent changes, so tools that reduce direct low-level tinkering may cut errors that make agent-produced documents unfit for client use, according to the release. This suggests that improving underlying libraries, not just prompts or AI skills, could be a meaningful lever for making agents reliable enough for professional document workflows in law, finance and consulting.
- Paper Office adds three Python packages to help agents edit Word, PowerPoint and Excel files more reliably.
- Testing showed higher success rates and far less direct XML-level editing compared to legacy libraries and Anthropic's Office skills.
- The release argues that basic software tooling, not just prompts or skills, is key to making agents fit for real professional document work.
Source: paperinstruments.com — Paper Instruments, 2026-09-24
Published there as: “Looks promising for document editing with your agent”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.