🔍 Read the full analysis: Can A Vibe-Coded PDF Editor Become Reliable? on ThorstenMeyerAI.com
Get smart everyday buys delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Thorsten Meyer says his AI-built PDF editor, Docestra, went from a one-day prototype to version 1.5.1 after testing and review exposed problems with saving, encrypted files and redaction. His account reports 745 unit tests and no leaks in a 917-PDF redaction test, but the app is private, the results come from its creator, and time or cost savings have not been measured.
Thorsten Meyer says his AI-built PDF editor, Docestra, moved from a first working version to version 1.5.1 in six days after tests on real files exposed problems with saving, encrypted PDFs and redaction. His account, published November 13, describes how testing and review changed the software, while noting that Docestra is private and not available to download.
Meyer says an AI coding agent built the first version from a one-sentence brief on November 6. The local web app could open PDFs, add highlights and notes, draw, sign and fill forms, rearrange pages, and save files. It was written in TypeScript using Anthropic’s Claude Code with Claude Opus, with OpenAI’s Codex reviewing the code, according to Meyer. The app runs in a browser on his Mac; he says files stay on the machine.
The first major problem emerged in a robustness run of 930 real PDFs. Meyer reports that a save operation produced a broken cross-reference table for a file created by a common PDF generator, causing markups to disappear from the saved document. He says he changed the save process to avoid compressed object streams. In a separate test, 52 of 69 encrypted PDFs would not allow annotations under restrictions set by their authors. Because the PDF library could not re-encrypt edited files, Meyer says Docestra was changed to require confirmation before unlocking and to save an unprotected copy without overwriting the original.
Redaction required a larger redesign. Meyer says an early “white-out” tool merely covered text, leaving the underlying content in the PDF. A first attempt at removing content in place drew 14 issues in one independent AI review and 13 in a second. He then switched to converting redacted pages into images, filling the marked areas in the pixels and adding a searchable text layer that omits text touching those areas. Meyer reports that this approach showed zero leaks in a test of 917 PDFs. He also says automated unit tests grew from 17 in the first version to 745 in version 1.5.1.
Docestra · Nov 6–13, 2026
From one prompt to a tool I can rely on
What the first version gave me, what real files broke, and what it took to fix it, from my project log.
What the tool rests on (schematic)
Built bottom up. The top layer is a plan, not a finished feature.
What happened, by date
- Nov 6One-sentence brief; first working version committed the same evening
- Nov 6Test over 930 real PDFs finds a save bug that dropped markups
- Nov 752 of 69 encrypted PDFs forbid annotations: unlock only into a new copy
- Nov 8Real redaction: review finds 14, then 13 problems; simpler design, 0 leaks in 917 PDFs
- Nov 12Version 1.5.1: 745 unit tests, 32 browser test suites
- Nov 13Version 2.0 starts: one shared PDF engine for our AI agents (in development)
Source: Docestra project records (development log, plans for versions 1 and 2.0, commit history), Nov 6–13, 2026. The layer stack is a schematic, not a measurement.
Graphic: Powered by Thorsten Meyer AI
Testing Exposed PDF Reliability Gaps
The account illustrates a distinction between a prototype that performs expected tasks and software that behaves safely across varied documents. A save bug can make annotations disappear; a cosmetic cover-up can leave sensitive text available underneath. Those are not simply interface defects when people use a PDF editor for contracts, forms or records.
Meyer’s experience suggests that the time required to test, review and revise an AI-generated tool can be substantial. He says that for several packages, testing and rework took longer than building. That is his experience with one private project, not a measured comparison of AI-assisted development costs or a general finding about all vibe-coded software. He also says he has not measured time or cost savings.
The reported test results may reassure users about the specific fixes Meyer describes, but they do not establish how Docestra performs on every PDF or in other environments. The app is not publicly available, so readers cannot independently test it or inspect its code based on the account.
Products Worth Considering
Vari Ergo Electric Standing Desk, 54×26, Light Wood/White
Samsung 32-Inch Odyssey G5 QHD 180Hz Gaming Monitor
Apple MacBook Neo 13-inch Laptop with A18 Pro Chip, 8GB Memory, 256GB SSD, Liquid Retina Display, and 1080p Camera — Blush
As an affiliate, we earn on qualifying purchases.
From One-Day Prototype to Version 1.5.1
Meyer uses “vibe coding” to describe building software mainly by giving an AI model plain-language instructions rather than writing the code himself. He says Docestra’s first working version was committed at 9:47 p.m. Central European Time on November 6, 2026. It already included a broad set of viewing, annotation and page-management features.
The subsequent revisions addressed problems that did not show up in the initial working version. Alongside saving and encryption, Meyer says licensing rules shaped the project: he did not ship fonts included with a PDF viewer library because they were under GPL-2.0, which his project rules excluded. He says he later checked the licences of libraries added for text recognition and text editing.
By November 12, Docestra had reached version 1.5.1. Meyer says work on version 2.0 began November 13 and remains unfinished. His account is a first-person report about one tool’s development, rather than an external audit or an announcement that the software is ready for wider release.
““Vibe coding gets you a working tool quickly. Turning it into a tool you can rely on is the real work.””
— Thorsten Meyer
Independent Verification Is Still Missing
The test counts and outcomes are reported by Meyer; the source material does not provide an independent audit, test files or detailed procedures that would let readers reproduce the results. The reported zero leaks in 917 PDFs applies to that test set and design, not necessarily to every document, redaction pattern or future version.
It is also unclear how the editor handles the full range of PDF formats and security settings beyond the cases described, or whether other reviewers would reach the same conclusions. Docestra is private and unavailable to download, and Meyer says version 2.0 is unfinished. The account does not report a public release date or independent measurements of reliability, time saved or cost.
Version 2.0 Remains in Development
Meyer says work on Docestra version 2.0 began on November 13, 2026, but he does not give a release date or describe a completed feature set. For now, the next development step is ongoing work by the creator; the source material does not identify a public launch, external testing program or independent review planned for version 2.0.
For readers considering AI-built tools, the account’s practical next step is to test the software against real files and keep records of problems and fixes. That is not a substitute for independent security or reliability testing, particularly for sensitive documents. Whether Docestra’s reported improvements hold up beyond Meyer’s test sets remains unverified.
Key Questions
What is Docestra?
Docestra is a PDF editor that Thorsten Meyer says he built with AI coding agents. It runs as a local web app on his Mac and, according to Meyer, keeps files on the machine. It is private and not available to download.
What problems did testing find?
Meyer reports that a save operation caused markups to disappear from one real PDF, that many tested encrypted PDFs did not permit annotations under their authors’ restrictions, and that an early white-out feature covered text without removing it.
How did Meyer change the redaction feature?
After an initial in-place approach received repeated review findings, Meyer says he changed to a design that turns redacted pages into images, fills the marked areas in the pixels and adds a searchable text layer without text touching the marks. He reports no leaks in a test of 917 PDFs; that result has not been independently verified in the source material.
Is Docestra publicly available or independently verified?
No. Meyer describes it as private and not available to download. The test results in his account are his own reported findings; the source material does not document an independent audit.
Has Meyer shown that vibe coding saves time or money?
No. Meyer says he has not measured time or cost savings. He reports that testing and rework took longer than building for several software packages, but that observation is not a project-wide or industry-wide cost comparison.
Source: ThorstenMeyerAI.com
