🔍 Read the full analysis: Why The Finishing Work On My AI PDF Editor Was Worth Doing on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer says his AI coding agent produced a working PDF editor, Docestra, in one day, but testing real documents exposed save, encryption and redaction problems. He reports that six days of revisions, independent AI review and expanded tests brought the private app to version 1.5.1; time and cost savings have not been measured.
Thorsten Meyer says he spent six days refining Docestra, a private PDF editor first built with an AI coding agent, after tests on real documents exposed problems that the initial version had not revealed. In an account published November 13, Meyer said the work led to version 1.5.1 on November 12, including changes to saving, encrypted files and redaction; the tool is not available to download.
Meyer said he gave an AI coding agent a one-sentence brief on November 6 to build a PDF editor for his own use. A working version was committed that evening at 21:47 Central European Time. It could open PDFs, add highlights and notes, draw and sign, fill forms, rearrange pages and save files. Meyer described the app as a local TypeScript web application running in a browser on his Mac, with files staying on that machine.
Testing a collection of 930 real PDFs found a saving defect: on some files, the library could produce a broken cross-reference table, and markups disappeared from the saved document. Meyer said he changed the saving approach to avoid compressed object streams. In a separate test, 52 of 69 encrypted PDFs did not permit annotations under restrictions set by their authors. He said the library could not re-encrypt edited files, so Docestra now asks for confirmation before unlocking a file and saves an unprotected copy without overwriting the original. Meyer reported that all 69 test files then unlocked.
Redaction required a larger design change. The original white-out feature only covered content; text beneath the overlay remained in the PDF. Meyer said an independent AI review identified 14 problems in the first review round and 13 in the second of an initial true-redaction design. He then replaced that approach with one that converts pages containing redactions into images, applies the redactions to the pixels and adds a searchable text layer that omits text touching marked areas. He reported zero leaks in a test of 917 PDFs. These results are Meyer’s account of his 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
Why Reliability Took More Work
Meyer’s account illustrates a distinction relevant to people using AI-generated software: a tool can appear functional on a few sample files and still fail on documents encountered in ordinary work. A save defect that drops annotations, or a visual cover-up that leaves sensitive text intact, could undermine confidence in a PDF editor even if its basic controls work.
The reported changes address different risks. Testing uncovered a file-integrity problem; the encrypted-file workflow makes a trade-off between editing and preserving password protection; and the redaction redesign aims to remove marked content rather than merely conceal it. Meyer says he chose to keep developing Docestra because it fit his workflow and because fixes could be turned into tests. He reports that automated unit tests rose from 17 in the first version to 745 in version 1.5.1. That count describes the project’s tests, not a general measure of software quality.
There is no reported comparison of development time or cost against conventional software development, and Meyer says he has not measured savings. The account is a personal report about one private project, rather than evidence that AI-built applications are generally reliable or that the same methods will produce similar results elsewhere.
Popular right now
eufy E25 Robot Vacuum and Mop with HydroJet Roller Mop, 20,000 Pa Suction
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Cypress Green
De'Longhi La Specialista Touch Semi-Automatic Espresso Machine with Grinder
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 functions in plain language to an AI model, which then writes the code. He says Anthropic’s Claude Code, running Claude Opus, produced Docestra’s code, while OpenAI’s Codex reviewed it from the first day. The initial version included viewing, searching, markup, page management, combining and extracting pages, printing, undo and saving.
Features and safeguards developed over the following days as Meyer tried more documents and used the editor in his workflow. He also described checking software licences before including libraries and said the PDF viewer library’s GPL-2.0 fonts were not shipped with Docestra under his project’s rules. He wrote that testing and rework took longer than building some packages. The account was published November 13, the day after version 1.5.1; Meyer said version 2.0 was in development but unfinished.
“Vibe coding gets you a working tool quickly. Turning it into a tool you can rely on is the real work.”
— Thorsten Meyer, describing his conclusion about AI-assisted development
What the Tests Do Not Establish
The reported test results come from Meyer’s own work, and the source does not provide a third-party audit of the 930-PDF robustness run or the 917-PDF redaction test. It also does not describe the test sets in enough detail to establish how representative they are of PDFs used by other people. The zero-leak result applies to the reported test, not to every possible document or redaction.
Docestra remains private and is not available to download, so outside users cannot independently assess its behavior. Meyer reports no measured time or cost savings, and the account does not compare the project’s results with a conventionally developed editor. Version 2.0 was still unfinished when he published his report.
Version 2.0 Remains in Development
Meyer said development of Docestra 2.0 began November 13, 2026, but he did not give a release date or describe a finished feature set. He said he wanted to add text recognition and true text editing, with library licences reviewed before packages were included. The next stated milestone is completion of that work; public availability has not been announced.
Until then, the available account is a report on the private version 1.5.1 and its tests. Meyer has not published a plan to release the app, provide an external audit or measure time and cost savings.
Key Questions
What is Docestra?
Docestra is a PDF editor Meyer says he built for personal use with AI coding agents. It runs locally in a browser on his Mac, and he says files stay on the machine. It is not available to download.
What problems did testing reveal?
Meyer reported a save issue that caused markups to disappear from some files, limits on annotating encrypted PDFs, and a first redaction tool that covered text without removing it from the PDF.
How did the redaction feature change?
After two AI review rounds found issues in the first true-redaction design, Meyer said he changed the method. Pages with redactions are converted to images, marked areas are applied to the pixels, and a searchable text layer omits text that touches those areas. He reported zero leaks in a test of 917 PDFs.
Has Meyer shown that Docestra saves time or money?
No. Meyer explicitly said he has not measured time or cost savings. His account describes the development process and test results, not a comparison with other ways of building or buying software.
When will Docestra 2.0 be released?
No release date was provided. Meyer said version 2.0 was in development as of November 13, 2026, and was not finished.
Source: ThorstenMeyerAI.com
