🔍 Read the full analysis: How I Went From Vibe-Coding To A More Reliable PDF Editor on ThorstenMeyerAI.com
TL;DR
Thorsten Meyer says an AI coding agent produced a working first version of his private PDF editor, Docestra, on Nov. 6, 2026. Testing real files over the next six days exposed save, encryption and redaction issues; by version 1.5.1, the project had grown from 17 to 745 automated unit tests. Meyer has not measured time or cost savings, and the app is not available to download.
Thorsten Meyer says an AI coding agent built the first working version of his PDF editor, Docestra, on November 6, 2026; six days of testing then exposed problems with saved markups, encrypted files and redaction. In an account published November 13, Meyer said the private, local web app reached version 1.5.1 on November 12, illustrating how much testing and revision his prototype needed before he considered it more reliable.
Meyer said he gave an AI coding agent a one-sentence request to build an Acrobat-like tool. The first version could open and search PDFs, add highlights and notes, draw, sign and fill forms, rearrange pages, and save files. The app was built in TypeScript and runs in a browser on his Mac; Meyer said the files remain on the machine. He used Anthropic’s Claude Code with Claude Opus to generate code and had OpenAI’s Codex review it from the start.
A test run on 930 real PDFs found a save problem: for a file produced by a common PDF generator, the PDF library could create a broken cross-reference table, causing markups to disappear from the saved document. Meyer said he changed the save process to avoid compressed object streams. Tests also found that 52 of 69 encrypted PDFs did not allow annotations under restrictions set by their authors. Because the library could not re-encrypt files after editing, Meyer chose to ask users to confirm an unlock and save an unprotected copy, without overwriting the original. He said all 69 test files then unlocked.
Redaction required a more substantial redesign. The first version’s “white-out” tool only covered text visually; the underlying content remained in the PDF. Meyer said an independent AI review identified 14 issues in the first round of review of his initial real-redaction design, followed by 13 in the second. He replaced that approach with one that turns marked pages into images, removes the marked content from the pixels and adds searchable text that excludes text touching a redaction. A test on 917 PDFs found no leaks, according to Meyer; that result describes his test set and is not a guarantee about other documents.
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 File Testing Changed the Build
The account highlights the gap between a working demonstration and software that handles varied documents safely. A save bug can erase edits without an obvious warning, while a visual cover-up can leave sensitive text in a file. Those risks matter to people using document tools for contracts, forms or other records, even when the tool is built for personal use.
Meyer’s experience also shows why review and regression tests are part of development rather than a final polish. He said the project’s 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. That is a measure of the project’s test suite, not proof that every possible PDF issue has been eliminated. Meyer said he has not measured time or cost savings, so the account does not establish that the AI-assisted process was cheaper or faster overall.
Popular right now
Philips 2200 Series Fully Automatic Espresso Machine with Classic Milk Frother
Ninja AF181 Air Fryer XL, 6.5 Qt, 6-in-1, Dark Grey
Dyson V8 Cyclone Cordless Vacuum
As an affiliate, we earn on qualifying purchases.
From One-Day Prototype to Version 1.5.1
Meyer describes “vibe coding” as building software by telling an AI model what to make in plain language and having it write the code. Docestra began as a personal project on November 6, and the first working version was committed at 21:47 Central European Time, according to his account. The initial feature list was already broad, including page rotation and deletion, combining or extracting pages, printing and undo.
But the first working build was tested on a limited set of files, Meyer said. The robustness run, encryption decisions and redaction reviews followed as he tried the editor against real documents. He also said he excluded bundled fonts distributed under the GPL-2.0 licence because his project rules barred GPL code, and later checked licences before adding libraries for text recognition and editing. His notes, he said, record testing and rework taking longer than building for several packages.
„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 Tests
The reported test results come from Meyer’s own project and test sets. The account does not provide independent verification of the 930-file robustness run or the 917-file redaction test, nor does it detail every PDF type, configuration or failure condition covered. A zero-leak result in that test set does not establish that redaction is error-free in all cases.
Docestra is private and, Meyer said, is not available to download. He has not measured time or cost savings, and it is unclear when version 2.0 will be finished or what features it will contain. The account does not report external users’ experiences or an independent security assessment of the application.
Version 2.0 Remains in Development
Meyer said work on Docestra version 2.0 began on November 13, 2026, the publication date of his account, and that it is not finished. He did not give a release date or describe a planned public launch. For now, the next reported milestone is continued development of the private tool; readers should not treat the project as a downloadable product or the reported tests as a certification of safety.
Key Questions
What is Docestra?
Docestra is Meyer’s private PDF editor, built as a local web app in TypeScript. He said it runs in a browser on his Mac and that files do not leave the machine.
What problems did testing uncover?
Meyer reported a save bug that could make markups disappear, restrictions affecting annotations on some encrypted PDFs, and a first redaction tool that covered text without removing it from the file.
Does the redaction test prove Docestra is secure?
No. Meyer said the redesigned approach produced no leaks in a test of 917 PDFs. That is a reported result from a specific test set, not proof that every document or use case is free of redaction risks.
Can the public download Docestra?
No. Meyer described Docestra as a private tool that is not available to download. He said version 2.0 was in development as of November 13, 2026, but gave no release date.
Did AI make the project faster or cheaper?
The first working version was committed on the day Meyer gave the coding agent its brief, but he said he has not measured time or cost savings. His account also describes testing and rework that, for some packages, took longer than building.
Source: ThorstenMeyerAI.com
