🔍 Read the full analysis: Why Reliability Mattered More Than Vibe-Coding My PDF Editor 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 produced a usable first version of his private PDF editor, Docestra, in one day. Six days of testing and review exposed issues with saving, encrypted files and redaction; by version 1.5.1, the project had grown to 745 automated unit tests. Meyer has not measured time or cost savings, and Docestra 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 real documents over the following six days uncovered problems with saving, encrypted files and redaction. In an account published November 13, Meyer said version 1.5.1 had 745 automated unit tests, up from 17 in the first version, while stressing that the tool is private and its time and cost savings have not been measured.
Meyer said he gave an AI coding agent a one-sentence brief on November 6 to build an Acrobat-like tool for his own use. The first working version was committed that evening. It could open PDFs and support tasks such as highlighting, drawing, adding notes, signing, filling forms, rearranging pages and saving documents. Docestra is a local web app that Meyer says runs in a browser on his Mac, with files kept on the machine.
A test run across 930 real PDFs revealed that a PDF library could save some files with a broken cross-reference table, causing markups to disappear. Meyer said he changed the save process to avoid compressed object streams. The same testing found that 52 of 69 encrypted files did not permit annotations under their authors’ restrictions. Because the library could not re-encrypt edited files, Docestra was changed to request confirmation before unlocking a file and to save an unprotected copy without overwriting the original.
Redaction required a separate redesign. An early “white-out” tool covered text without removing it. After Meyer added a redaction feature that removed content in place, an independent AI review identified 14 problems in one round and 13 in a second. Meyer said he replaced that approach with one that rasterizes pages containing redactions and adds a searchable text layer that omits lines touching marked areas. He reported zero leaks in tests across 917 PDFs. The figures describe his own testing; the article does not provide independent verification.
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 Exposed Document Risks
The account illustrates why a tool that appears to work in a short demonstration may still fail on documents outside that small sample. A save defect can remove annotations, while a visual cover-up that leaves text in a PDF could expose information a user believes has been removed. These are practical risks for people handling contracts, forms or other sensitive records.
Meyer’s experience also shows that AI-generated code does not remove the need for decisions about privacy, file restrictions, licensing and testing. He said some packages required more time for testing and rework than for initial implementation. That is his account of one personal project, not a measured comparison of AI-assisted development with other methods. He has not reported measured productivity gains or financial savings.
Popular right now
Dyson V8 Motorhead Origin Cordless Vacuum
Philips Baristina Espresso Machine with Milk Frother (BAR308/60)
Vitamix VX1 Professional-Grade Blender, 64 oz, Black
As an affiliate, we earn on qualifying purchases.
From One-Day Build to Version 1.5.1
Meyer defines “vibe coding” as describing software in plain language to an AI model and having it write the code. He said Claude Code, running Claude Opus, produced Docestra’s code, while OpenAI’s Codex reviewed it from the beginning. Development moved from the initial version on November 6 to version 1.5.1 on November 12.
The editor’s feature set expanded beyond viewing and markup to include page rotation, deletion, combining and extracting pages, printing and undo. Meyer also said he excluded fonts shipped 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. Those details describe his project’s choices, not a general legal assessment of the libraries or their use.
“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 testing figures come from Meyer’s account, and no independent audit or test data is included in the source material. It does not identify the full composition of the PDF test sets, the precise methods used to check for redaction leaks, or how results might differ on other files and systems. The reported zero leaks applies to 917 PDFs in Meyer’s test, not all PDFs.
Docestra is not available to download, so readers cannot independently examine or reproduce the reported results from a public release. Meyer also said the tool’s time and cost savings have not been measured. His next major release, version 2.0, was still in development as of November 13.
Version 2.0 Remains in Development
Meyer said work on Docestra 2.0 began November 13 and that it was not finished when he published his account. He did not give a release date or say that the software would become publicly available. The next developments are therefore still within a private project, based on the information provided.
For now, the reported milestone is version 1.5.1 and its expanded test suite. Any claims about wider reliability, user adoption or measurable savings would require further evidence, including details or independent testing beyond Meyer’s account.
Key Questions
What is Docestra?
Docestra is a PDF editor Meyer built for his own use with help from AI coding agents. He says it runs locally in a browser on his Mac and is not available to download.
What problems did testing find?
Meyer reported a save issue that could make markups disappear, restrictions affecting annotations in 52 of 69 encrypted PDFs, and shortcomings in an early redaction method that covered text without removing it.
How did the redaction design change?
After an independent AI review flagged problems in two rounds, Meyer said he replaced in-place removal with a method that turns redacted pages into images and adds a searchable text layer without lines touching marked areas. He reported no leaks in tests of 917 PDFs; that result has not been independently verified in the supplied account.
Did Docestra save time or money?
That has not been measured. Meyer said testing and rework took longer than building for several packages, but did not provide a project-wide comparison of development time or costs.
When will Docestra 2.0 be released?
Meyer said version 2.0 was in development as of November 13, 2026, but did not give a release date. He also said Docestra is private and not available to download.
Source: ThorstenMeyerAI.com
