🔍 Read the full analysis: I Thought Coding The PDF Editor Was The Work—Then Came Reliability on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer says an AI coding agent built the first working version of his private PDF editor, Docestra, in one day. In tests and reviews over the following six days, he found problems involving saved annotations, encrypted files and redaction; he says version 1.5.1 had 745 unit tests. He has not measured time or cost savings, and the app is not available to download.
Thorsten Meyer says he used an AI coding agent to build a working PDF editor, Docestra, in one day, then spent six days testing and revising it after real documents exposed problems. In an account published November 13, Meyer reported fixes and design changes involving saved annotations, encrypted PDFs and redaction, arguing that a quick prototype is not the same as software users can rely on.
Meyer said he gave an AI coding agent a one-sentence brief on November 6, 2026, to build a PDF editor for his own use. He said the first version was committed at 9:47 p.m. Central European Time and could open PDFs, add annotations and signatures, fill forms, reorder pages and save files. The local TypeScript web app was built with Anthropic’s Claude Code running Claude Opus; Meyer said OpenAI’s Codex reviewed the code from the start.
The initial version appeared to work on the files Meyer had tried, but a run over 930 real PDFs exposed a save problem: a PDF library could produce a broken cross-reference table for a file made by a common PDF generator, causing markups to disappear from the saved document. Meyer said he changed the save process to avoid compressed object streams. The test also found that 52 of 69 encrypted PDFs did not allow annotations under restrictions set by their authors. Meyer said the library could not re-encrypt files after editing, so Docestra was changed to ask for confirmation, identify the restrictions and save an unprotected copy without overwriting the original.
Redaction raised a separate risk. Meyer said the original “white-out” feature covered text without removing it from the PDF. After he built a redaction feature on November 8, an independent AI review identified 14 issues in its first round and 13 in the second. Rather than continue patching that design, he said he replaced it with one that converts redacted pages to images and adds a searchable text layer that omits text touching the redacted areas. Meyer reported that a test on 917 PDFs found no leaks; the account does not describe the test methodology or establish how the feature performs on every document.
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
Why PDF Reliability Takes More Than a Demo
The problems Meyer describes show why a tool that appears functional on a handful of sample documents may still fail in ordinary use. A save defect can remove annotations, while a visual cover-up that leaves underlying text intact can expose information a user believes has been removed. Those are practical risks for anyone using a PDF editor for work, particularly when documents include sensitive details or need to be shared.
Meyer’s account also illustrates a limit of treating AI-generated code as a finished product. The model produced a broad set of features quickly, but testing, review and decisions about how the editor should handle restrictions took additional work. Meyer said automated unit tests rose from 17 in the first version to 745 in version 1.5.1, with tests added for problems found during review so that earlier defects could be caught if they returned.
The report is a developer’s account of one private application, not an independent evaluation of AI coding tools generally. Meyer said he has not measured time or cost savings. He also said some packages took longer to test and rework than to build, so the account does not establish that AI development made the overall project faster or cheaper.
Products Worth Considering
Dell UltraSharp U2724DE 27-Inch WQHD IPS Black Monitor
Dell UltraSharp U2424H 24-Inch FHD Monitor
Samsung 32-Inch Movingstyle Essential 4K UHD Smart Monitor with Adjustable Rolling Stand
As an affiliate, we earn on qualifying purchases.
From One-Day Build to Version 1.5.1
Meyer defines “vibe coding” as describing desired software to an AI model in plain language and having it write the code. Docestra began as a personal, local web app running in a browser on Meyer’s Mac; he said the files stay on that machine. Its first version included viewing, searching, highlighting, drawing, notes, signatures, form filling, page rotation and reordering, combining and extracting pages, printing, undo and saving.
The development account covers releases from version 1 on November 6 to version 1.5.1 on November 12. Alongside reliability fixes, Meyer said he reviewed software licences before adding libraries, citing a decision not to ship fonts covered by GPL-2.0 because his project rules exclude GPL code. He described later approvals for libraries intended to support text recognition and editing.
Meyer said he kept improving Docestra because it fit his own workflow and because tests and reviews could accumulate as the application changed. He described the editor as a private tool, not a released product, and said version 2.0 had been in development since November 13.
“Vibe coding gets you a working tool quickly. Turning it into a tool you can rely on is the real work.”
— Thorsten Meyer, writing on ThorstenMeyerAI.com
Limits of the Reported Test Results
The reported results come from Meyer’s own testing and are not presented as an independent audit of Docestra. His account says the redaction test found no leaks across 917 PDFs, but does not provide enough detail to establish the test’s coverage, methods or performance on other documents. It also does not say whether an external security assessment has been conducted.
Docestra remains private and is not available to download, according to Meyer, so readers cannot independently reproduce his tests from a public release. He has not reported measured time or cost savings, and the account does not establish how the editor performs for other users, on other systems or across a broader range of PDF files.
Version 2.0 Remains in Development
Meyer said work on Docestra version 2.0 began November 13, the date of his account, and that it was not finished. He did not give a release date or say that the application would become publicly available. The next stated step is continued development; no further milestones, independent test results or release plans were provided.
Key Questions
What happened to Docestra?
Thorsten Meyer says an AI coding agent produced the first working version of his private PDF editor on November 6, 2026. He then revised it through version 1.5.1 on November 12 after testing and review exposed reliability problems.
What problems did testing find?
Meyer reported a save issue that could make annotations disappear, restrictions affecting edits to encrypted PDFs, and weaknesses in an early redaction approach that covered text without removing it. He said he changed the save process and the way encrypted documents and redactions were handled.
Did the redaction test prove the feature is secure?
No. Meyer said a test across 917 PDFs found no leaks, but his account does not give the full methodology or establish results for every document or use case. The result should be treated as a reported test finding, not a general security guarantee.
Can people download Docestra?
Meyer said Docestra is private and not available to download. He did not announce a public release plan.
Did AI development save time or money?
Meyer said he has not measured time or cost savings. He reported that testing and rework took longer than building for several packages, so the account does not quantify the project’s overall efficiency.
Source: ThorstenMeyerAI.com
