🔍 Read the full analysis: From AI-Generated Code To A PDF Editor That Works Reliably 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 an AI coding agent produced the first version of his PDF editor, Docestra, in one day. Tests on real files and an independent AI review later exposed reliability and redaction problems, leading to changes across six days of development. Docestra is a private tool, and its developer has not measured time or cost savings.
Thorsten Meyer says his AI coding agent built the first working version of Docestra, a PDF editor, on November 6, 2026, but tests on real files quickly exposed problems that a successful demo had missed. In an account published November 13, Meyer describes six days of fixes and redesigns, including a save bug and a redaction feature that required a change in approach before it passed his tests.
Meyer says he gave an AI coding agent a one-sentence request for an Acrobat-like editor he could use. The first version, committed at 9:47 p.m. Central European Time that day, could open and search PDFs, add highlights, drawings, notes and signatures, fill forms, rearrange pages and save files. He describes it as a local TypeScript web app running in a browser on his Mac; according to Meyer, its files stay on that machine.
The first major problem surfaced in a robustness run over 930 real PDFs. Meyer says a PDF library could save one kind of file with a broken cross-reference table, causing markups to disappear from the saved document. He says he changed the save process to avoid compressed object streams. The same test set included 69 encrypted PDFs: according to his account, 52 did not permit annotations under restrictions set by their authors, and the library could not re-encrypt edited files. Docestra was changed to require confirmation before unlocking such a file and to save an unprotected copy without overwriting the original.
Redaction proved more difficult. An early “white-out” tool covered text visually but left the underlying content in the PDF, a limitation Meyer says was stated in the tool’s plan. A later in-place redaction design drew 14 problems in the first independent AI review and 13 in the second. Meyer then replaced it with a method that converts pages with redactions into images, removes marked material from those pixels and adds a searchable text layer that omits text touching a marked area. He reports that tests on 917 PDFs found no leaks; the result is his test finding, not an independent certification.
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 Real-File Testing Changed Docestra
The development account shows the gap between software that works in a demonstration and software that can be trusted with everyday documents. A save defect may not be obvious in the editor itself: users can see an annotation on screen and still lose it when they reopen the saved file. Encryption and redaction carry higher stakes because a user may believe a document remains protected or that sensitive information has been removed when neither is true.
Meyer’s experience also illustrates that generating code is only one part of building a dependable tool. Testing against varied documents, examining how libraries handle restrictions, and deciding when to replace a flawed design all took additional work. He says notes for several packages show testing and rework taking longer than implementation. That is his experience with this project, not a general measurement of AI coding tools.
There are limits to what readers can infer. Meyer says Docestra is private and unavailable to download, and he has not measured time or cost savings. The account therefore documents one developer’s process rather than providing a public product evaluation or evidence that the same approach will perform similarly elsewhere.
Products Worth Considering
Canon SELPHY CP1500 Compact Photo Printer Bundle with Ink and Paper Set, White
HP OfficeJet 250 Wireless Mobile All-in-One Inkjet Printer
LG 27US500-W 27-Inch 4K UHD IPS Monitor
As an affiliate, we earn on qualifying purchases.
From One-Day Build to Version 1.5.1
Meyer defines “vibe coding” as describing a software tool in plain language and having an AI model write much of the code. He says Claude Code, running Claude Opus, generated Docestra’s code, while OpenAI’s Codex reviewed it from the start. The initial editor included viewing, searching, annotations, form filling, page rotation and reordering, combining and extracting pages, printing, undo and saving.
The development timeline in Meyer’s account runs from the first version on November 6 to version 1.5.1 on November 12. Along the way, requirements grew out of his own use, including a way to work with encrypted PDFs. He also describes reviewing licences before adding libraries: the PDF viewer library includes fonts under GPL-2.0, which he says Docestra does not ship because his project rules exclude GPL code.
Meyer says the project’s automated unit tests increased from 17 in the first version to 745 in version 1.5.1. He says problems found during reviews were turned into tests intended to catch regressions. These figures and development details come from his published account; Docestra’s code and test suite are not available for outside inspection.
“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
What Docestra’s Tests Cannot Establish
The reported results come from Meyer’s own testing and have not been independently verified in the source material. The account does not provide the PDFs, test procedures or code needed for readers to reproduce the reported finding of zero redaction leaks across 917 files. The types of documents in each test set and how closely they represent broader real-world use are also not specified.
Docestra is not available to download, so outside users cannot assess its behavior or compare it with other PDF editors. Meyer also says he has not measured time or cost savings. His account does not establish whether the tool will meet a formal security standard, how it performs across other computers, or how its handling of encrypted files applies to every PDF restriction.
Version 2.0 Is Still in Development
Meyer says work on Docestra 2.0 began November 13, 2026, the date his account was published, and that the release is not finished. He mentions plans for text recognition and true text editing, with bundled libraries reviewed for their licences before inclusion. He does not give a release date or describe a completed public distribution plan.
The next stated step is continued development and testing. Until more details or a downloadable version are provided, Docestra’s performance beyond Meyer’s reported test sets remains unverified by outside users.
Key Questions
What is Docestra?
Docestra is a PDF editor that Thorsten Meyer says he built with AI coding agents for personal use. It runs as a local web app on his Mac and is not available to download.
What problems did testing find?
Meyer reports that a save-library issue caused markups to disappear from some saved files. He also found that many encrypted test PDFs restricted annotations and that an early white-out feature covered text without removing it from the PDF.
Did Docestra’s redaction feature pass independent testing?
Meyer says an independent AI reviewer identified 14 problems in the first redaction design and 13 in a second review. After he replaced the design, tests on 917 PDFs found no leaks, according to his account. The source does not describe an external security audit or provide materials for independent reproduction.
Is Docestra publicly available?
No. Meyer says the tool is private and not available to download. He has not announced a release date for version 2.0, which he says is still in development.
Has Meyer measured savings from using AI to build the editor?
No. Meyer explicitly says he has not measured time or cost savings. His account describes the development process and testing, but it does not quantify the project’s overall savings or compare them with another way of building the software.
Source: ThorstenMeyerAI.com
