🔍 Read the full analysis: The Best Part Of Building My Own PDF Editor Came After Coding 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 reports that an AI coding agent produced a working first version of his private PDF editor, Docestra, on Nov. 6, 2026. Testing hundreds of real files exposed save, encryption and redaction problems; by version 1.5.1 on Nov. 12, its automated unit tests had grown from 17 to 745. 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, in a day, but testing and revising it over the following six days exposed problems that a quick demonstration had missed. In an account published Nov. 13, Meyer said the private, locally run app reached version 1.5.1 on Nov. 12 after fixes involving PDF saving, encrypted files and redaction; he also reported that its automated unit tests grew from 17 to 745.
Meyer said he gave an AI coding agent a one-sentence brief on Nov. 6 to build an Acrobat-like editor he could use. The first version was committed at 9:47 p.m. Central European Time. It could open, search and save PDFs, add highlights, notes, drawings and signatures, fill forms, and reorder or combine pages. The local web app was written in TypeScript, runs in a browser on his Mac and, according to Meyer, keeps files on that machine.
The problems emerged when Meyer tested 930 real PDFs. He said a save operation could produce a broken cross-reference table in a file from a common PDF generator, causing markups to disappear. He changed the saving method to avoid compressed object streams. In a separate test, 52 of 69 encrypted PDFs did not permit annotations under restrictions set by their authors. Because the PDF library could not re-encrypt edited files, Meyer chose to require confirmation before unlocking them and to save an unprotected copy rather than overwrite the original. He reported that all 69 test files then unlocked.
Redaction required a larger design change. An early white-out tool covered text without removing it from the PDF, a limitation Meyer said the tool disclosed. After he built an initial feature intended to remove content, an independent AI review identified 14 problems in its first round and 13 in its second. Meyer said he abandoned that approach. The replacement converts pages with redactions into images, covers the marked areas in the pixels and adds searchable text that excludes lines touching a mark. In a test over 917 PDFs, he reported zero leaks. Those results are from his own tests, not an independently published evaluation.
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 Changed the Editor
Meyer’s account illustrates a distinction that matters to people using AI-generated software: a tool that works on a few sample documents may still fail on files encountered in daily work. Here, the risks were practical. A save defect could remove visible edits, while an ineffective redaction could leave sensitive text in a document that a user believes has been cleaned.
The reported revisions also show that the work did not end when the interface and core features appeared. Meyer says testing, review and rework sometimes took longer than building. His unit-test count increased from 17 to 745 by version 1.5.1, with tests added for problems found during review. That is a measure of testing effort, not proof that every failure has been found or that the editor is safe for every use.
The account is relevant to individuals and organizations experimenting with AI coding agents, especially when software handles sensitive documents. It does not establish that AI-built applications generally save time or money: Meyer explicitly says he has not measured either. Nor is Docestra currently offered for download, so readers cannot independently reproduce the reported results from the article alone.
Products Worth Considering
Logitech MX Master 4 Ergonomic Wireless Mouse with Haptic Feedback, Graphite
EcoFlow DELTA 3 Max Portable Power Station, 2048Wh LFP Battery
Bose SoundLink Max Portable Bluetooth Speaker, IP67 Waterproof, Black
As an affiliate, we earn on qualifying purchases.
From One-Day Build to Version 1.5.1
Meyer describes “vibe coding” as building software mainly by telling an AI model what to make in plain language and letting it write the code. He said Anthropic’s Claude Code, running Claude Opus, produced Docestra’s code, while OpenAI’s Codex reviewed it from the first day. The account is a developer’s description of his own project, rather than an independent product review or a published scientific study.
After the initial build on Nov. 6, 2026, Meyer says he worked through successive versions until 1.5.1 on Nov. 12. The changes included decisions about encrypted files and a replacement design for redaction. He also said he excluded fonts shipped under the GPL-2.0 licence because his project rules do not permit GPL code, and checked licences for later libraries before adding them.
Version 1.5.1 includes features beyond the initial editor, according to Meyer, but the source account centers on the gap between a working first build and a tool he felt able to rely on. Its central evidence is his testing and development record; the editor remains private and the claims have not been independently verified.
“Vibe coding gets you a working tool quickly. Turning it into a tool you can rely on is the real work.”
— Thorsten Meyer, describing the project
Limits of the Reported Tests
The reported test results come from Meyer; the source account does not provide an independent audit, test protocol or detailed breakdown of the PDF collections. It is not clear how the 930- and 917-file sets were selected, what kinds of documents they covered or how broadly the results apply to PDFs outside those sets. Meyer’s report of zero redaction leaks therefore describes that test run, not a guarantee that the feature cannot expose information.
It is also unclear whether further testing will uncover new failures. Meyer has not published measured time or cost savings, and Docestra is private and unavailable for download. His account does not establish how the editor would perform across other computers, operating systems or users’ workflows.
Version 2.0 Remains in Development
Meyer said development of version 2.0 began Nov. 13, 2026, the date of his account, and that it was not finished. He did not give a release date or say whether the software would become publicly available. He also said he wants text recognition and true text editing, subject to checking the licences of bundled libraries.
The next milestone is therefore continued development rather than a confirmed public release. Whether version 2.0 will add those features, how it will be tested and whether Meyer will publish enough detail for independent evaluation remain open.
Key Questions
What is Docestra?
Docestra is a private PDF editor Meyer built for his own use with AI coding agents. He says it runs as a local web app on his Mac, with files staying on the machine.
What did testing reveal?
Meyer said a run over 930 real PDFs found a save bug that could make markups disappear. He also reported restrictions on annotations in 52 of 69 encrypted test files and problems with the first redaction design.
Did the redaction feature pass independent testing?
No independent test result is provided. Meyer said an independent AI reviewer found problems in two rounds on the first design, after which he switched approaches. He reported zero leaks in a 917-PDF test of the replacement, but that result has not been independently verified.
Can readers download Docestra?
No. Meyer says the app is private and not available to download. He has not announced a public release.
Has Meyer shown that the AI-built editor saved time or money?
No. He says he has not measured time or cost savings. His account describes the build and subsequent testing, not a quantified comparison with other development methods.
Source: ThorstenMeyerAI.com
