🔍 Read the full analysis: I Vibe-Coded A PDF Editor—Then Put It Through The Reliability Work on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
Thorsten Meyer says an AI coding agent built the first working version of his private PDF editor, Docestra, in one day. Six days of testing real documents 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 six days of testing turned Docestra, his AI-built PDF editor, from a working first version into a more heavily tested tool, after real documents exposed problems with saving, encrypted PDFs and redaction. In an account published November 13, Meyer says a coding agent produced the first version on November 6, while testing and review continued through version 1.5.1 on November 12.
Meyer describes Docestra as a local web app written in TypeScript that runs in a browser on his Mac, with files staying on the machine. He says AI coding agents using Anthropic’s Claude Code and Claude Opus wrote the software, while OpenAI’s Codex reviewed it from the start. The first version included PDF viewing and search, annotations, signatures, form filling, page reordering and saving.
A test run over 930 real PDFs found 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 after saving. Meyer says he changed the save process to avoid compressed object streams. His account does not identify the PDF generator or provide independent test results.
Testing also found that 52 of 69 encrypted PDFs did not allow annotations under restrictions set by their authors. Meyer says the library could not re-encrypt edited files, so Docestra was changed to require confirmation before unlocking such documents and to save an unprotected copy without overwriting the original. He reports that all 69 test files then unlocked.
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’s Design
The account illustrates the difference between a software demonstration and a tool that can handle varied documents. Meyer says the first version worked on the files he initially tried, but a broader test run found a defect that could make saved annotations disappear. That matters to anyone relying on an editor to preserve changes, particularly when a file is part of a work process.
Redaction raised a more serious distinction. Docestra’s original “white-out” tool covered content visually but left the underlying text in the PDF, Meyer says. A user could mistake that appearance for removal of sensitive information. After an independent AI review identified 14 problems in the first review round and 13 in the second for an in-place redaction design, Meyer says he replaced it rather than continuing to patch it. The new approach converts marked pages to images and adds a searchable text layer that excludes lines touching a redaction; Meyer reports zero leaks in tests of 917 PDFs.
Those results are the author’s reported test outcomes, not a published independent audit. Still, the development account offers a practical warning: generating code quickly does not establish that it preserves documents, respects restrictions or removes sensitive content reliably.
Popular right now
Shark WANDVAC Power PET Cordless Handheld Vacuum, White, WV410WH
Hisense 75-Inch U7 Mini-LED ULED 4K UHD Smart Google TV (75U7SG)
Dyson Ball Animal 3 Upright Corded Vacuum
As an affiliate, we earn on qualifying purchases.
From One-Day Build to Version 1.5.1
Meyer defines vibe coding as building software mainly by describing desired features in plain language to an AI model, which then writes the code. He says Docestra’s first working version was committed at 9:47 p.m. Central European Time on November 6, the day he gave the agent a one-sentence brief to make an Acrobat-like tool for his use.
The project then went through releases from version 1 on November 6 to version 1.5.1 on November 12. Meyer says automated unit tests rose from 17 in the first version to 745 in version 1.5.1, with tests added for problems found during reviews so they could be caught if they returned. He also says license checks affected which libraries and fonts he included, and that later additions such as text recognition and text editing were reviewed for licensing.
Docestra is private and unavailable to download. Meyer says version 2.0 entered development on November 13 and remains unfinished. His account is a first-person report about one project, not a comparative study of AI-generated software or a general measure of reliability.
“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’s main lesson
Reported Tests, Limited Verification
The test counts and outcomes in the account are Meyer’s own reports. The source does not include the test corpus, test methods, the independent reviewer’s full findings or a third-party audit of the redaction process. It is not clear how the reported 917-PDF redaction test was designed, what counted as a leak, or how broadly its results apply to other documents.
It is also unclear how Docestra would perform across operating systems, PDF software libraries or documents outside the tested sets. Meyer says the application is private and has not measured time or cost savings, so the account does not establish whether this approach is more efficient than existing tools or suitable for other users.
Version 2.0 Is Still in Development
Meyer says version 2.0 began development on November 13, 2026, but is not finished. His account does not give a release date or say whether the application will become publicly available. For now, the next development milestone is an unfinished version of a private tool, rather than a product launch.
The reported changes point to the work still required: testing against varied PDFs, preserving user changes, handling document restrictions carefully, reviewing security-sensitive features and checking third-party licenses. Whether Docestra’s design and test results will be independently evaluated or shared more broadly remains unknown.
Key Questions
What is Docestra?
Docestra is a private PDF editor that Meyer says he built with AI coding agents. It runs as a local browser app on his Mac and is not available to download.
What did testing find?
Meyer says a run over 930 PDFs exposed a save bug that could make markups disappear. Tests also found limits with encrypted PDFs and showed that the original white-out tool covered text without removing it.
Does the redaction test prove the feature is secure?
No independent proof is provided in the account. Meyer reports zero leaks in tests of 917 PDFs using the revised design, but the test methods and results have not been independently verified in the supplied source.
Did the project save time or money?
Meyer has not measured time or cost savings. He says testing and rework took longer than building for several packages, so the account does not quantify the project’s overall efficiency.
What happens to Docestra next?
Meyer says version 2.0 is in development as of November 13, 2026, and is unfinished. He has not announced a release date or public download.
Source: ThorstenMeyerAI.com
