🔍 Read the full analysis: I Made My Own PDF Editor With AI. Reliability Was Worth The Effort on ThorstenMeyerAI.com
Get the latest gadgets 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 built the first working version of his private PDF editor, Docestra, on Nov. 6, 2026. Six days of testing and reviews exposed problems with saving, encrypted files and redaction; by version 1.5.1, the project had 745 unit tests. Meyer has not measured time or cost savings, and the tool is not available to download.
Thorsten Meyer says an AI coding agent produced the first working version of his PDF editor, Docestra, on Nov. 6, 2026, but six days of testing and review revealed problems that a basic demonstration had missed. His account, published Nov. 13, describes fixes for saved annotations, encrypted PDFs and redaction, and says the private app’s automated unit tests grew from 17 to 745 by version 1.5.1.
Meyer, writing on ThorstenMeyerAI.com, says he gave an AI coding agent a one-sentence request to build an Acrobat-like tool he could use. The initial version was committed at 9:47 p.m. Central European Time that day. It could open PDFs, add highlights and notes, draw, sign, 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, files stay on the machine.
The first major issue surfaced in a robustness run across 930 real PDFs. Meyer says a PDF library could save a file with a broken cross-reference table, causing markups to disappear. He changed the save process to avoid compressed object streams. In a separate test, 52 of 69 encrypted PDFs did not permit annotations under restrictions set by their authors. Meyer says the library could not re-encrypt edited files, so Docestra asks for confirmation before unlocking such a file and saves an unprotected copy without overwriting the original. He reports that all 69 test files then unlocked.
Redaction required a different design. The original “white-out” feature covered text without removing it, a distinction Meyer says was stated in the tool but could be misunderstood. An independent AI review identified 14 problems in the first real-redaction design and 13 in a second review. Meyer says he stopped patching that approach and instead converted redacted pages to images, applied marks to the pixels and added a searchable text layer excluding lines that touched a redaction. He reports that tests on 917 PDFs found no leaks. These are results reported by the developer, not an independent public audit.
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
Meyer’s account illustrates a gap between producing a functional prototype and establishing that a document tool behaves safely across varied files. A missing markup can undermine ordinary editing, while a redaction that only hides text can leave sensitive information in a document. Encryption and software licensing also impose decisions that a short feature brief may not address.
The practical lesson is not that AI-generated software is inherently unreliable, nor that this particular app is ready for broad use. It is that claims of usefulness depend on testing against real inputs, reviewing consequential features and retaining checks against regressions. Meyer says each issue found in review now has a test intended to catch its return. He also says he has not measured time or cost savings, so his experience does not establish that AI coding made the whole development process faster or cheaper.
Products Worth Considering
Bose TV Speaker (2nd Gen) Compact Soundbar
DJI Osmo Mobile 8 3-Axis Phone Gimbal Stabilizer
Amazon Kindle Paperwhite (12th Generation, 2024), 16GB, 7-Inch Display, Raspberry
As an affiliate, we earn on qualifying purchases.
From One-Day Prototype to Version 1.5.1
Meyer defines “vibe coding” as building software by describing desired behavior in plain language and having an AI model write the code. He says Docestra’s code was produced using Anthropic’s Claude Code with Claude Opus, while OpenAI’s Codex reviewed it from the first day. The initial feature set included viewing and searching, annotations, form filling, page operations, combining and extracting pages, printing, undo and saving.
Development continued through successive versions from Nov. 6 to version 1.5.1 on Nov. 12. Meyer also says the PDF viewer library included fonts under the GPL-2.0 license, which his project rules excluded, so Docestra does not ship those fonts. He later says he explicitly checked the licenses of bundled libraries added for text recognition and text editing. The project is described as a personal, private tool, not a downloadable product.
“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 the Tests Do Not Establish
The reported test results come from Meyer’s own account; the source does not provide an independent audit, the full test suite or detailed methods for reproducing the results. The reported zero leaks across 917 PDFs applies to that test set and does not establish that every redaction in every file is secure. It is also unclear how Docestra handles PDF types or workflows outside the examples Meyer describes.
There is no measured comparison of development time, cost or reliability against a commercial editor, and the app is not available to download for outside evaluation. Meyer says version 2.0 entered development on Nov. 13 and remains unfinished. Its scope, release date and further test results have not been provided.
Version 2.0 Remains in Development
Meyer says work on Docestra 2.0 began Nov. 13, 2026, the date of his account, but that the version is not finished. He has not announced a release date or said whether the app will become publicly available. For now, the next developments to watch are whether further testing leads to changes, what version 2.0 includes and whether Meyer publishes evidence that allows others to evaluate the reported results.
Key Questions
What is Docestra?
Docestra is Meyer’s private PDF editor, described as a local web app running in a browser on his Mac. Its features include annotations, signatures, form filling, page rearrangement and saving.
How long did it take to build?
Meyer says the first working version was committed on Nov. 6, 2026, the day he gave the AI coding agent its brief. He then spent six days testing and revising the tool through version 1.5.1 on Nov. 12; he has not measured total time or cost savings.
Was the redaction test independently verified?
The source says Meyer’s redaction design received reviews from an independent AI reviewer and that tests across 917 PDFs found zero leaks. It does not describe an independent external audit of those results, so the finding should be treated as Meyer’s reported test result.
Can the public download Docestra?
No. Meyer describes Docestra as private and says it is not available to download.
What is known about Docestra 2.0?
Meyer says version 2.0 has been in development since Nov. 13, 2026, and is unfinished. He has not stated a release date or detailed its planned features.
Source: ThorstenMeyerAI.com
