I Used AI To Build A PDF Editor; Reliability Took More Work
AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

Thorsten Meyer says an AI coding agent produced a working first version of his private PDF editor, Docestra, on Nov. 6, 2026. Testing real documents exposed problems with saved annotations, encrypted files and redaction; by Nov. 12, version 1.5.1 had 745 unit tests and 32 browser test suites, according to his project records.

Thorsten Meyer says he built the first working version of Docestra, a private PDF editor, with an AI coding agent in one day, then spent six days addressing failures found in real files. His project records say testing exposed a save bug that could make annotations disappear, restrictions in encrypted PDFs, and flaws in an early redaction design—problems that matter because a PDF tool must preserve changes and handle sensitive content reliably.

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

Meyer’s account says he gave an AI coding agent a one-sentence brief on Nov. 6, 2026, to build an Acrobat-style tool for his own use. The first version was committed at 21:47 Central European Time. It could open PDFs, add highlights and notes, draw, sign, fill forms, rearrange pages and save. The local TypeScript web app runs on his Mac in a browser; Meyer says files stay on the machine. He says Anthropic’s Claude Code, running Claude Opus, wrote code, while OpenAI’s Codex reviewed it from the first day.

A test run across 930 real PDFs found that saving could produce a broken cross-reference table in a file made by a common PDF generator, causing markups to disappear. Meyer says he changed the save approach to avoid compressed object streams. Testing also found that 52 of 69 encrypted PDFs prohibited annotations under restrictions set by their authors. Because the PDF library could not re-encrypt files after editing, he chose to unlock those files only after a confirmation and save an unprotected copy, leaving the original untouched. He reports that all 69 test files then unlocked.

The first version’s white-out tool covered text without removing it. Meyer says he added a true redaction feature on Nov. 8, but an independent AI review identified 14 problems in its first review and 13 in a second, mainly gaps in earlier fixes. He says he stopped patching the initial design and switched to a simpler one, which had no reported leaks across 917 PDFs. By Nov. 12, version 1.5.1 had 745 unit tests and 32 browser test suites, according to his project log. Version 2.0 development began Nov. 13.

At a glance
reportWhen: Development described for Nov. 6–13, 20…
The developmentA developer’s six-day project log describes how real-file testing and review changed an AI-built PDF editor after its first version appeared to work.
Screenshot of the Docestra PDF editor, version 1.5.1, in a browser window: a top bar with open, save, undo and zoom controls, a row of markup tools and colour swatches, a pages panel on the left and a sample document titled Sample Project Brief in the middle. Notes: runs locally in the browser on my Mac, files stay on the machine; sample PDF made for this screenshot
Docestra 1.5.1 with a sample document made for this screenshot. Own screenshot, November 13, 2026. Graphic: Thorsten Meyer AI.

Reliability Depends on Real Documents

The account illustrates a gap between producing a working demonstration and handling documents safely across varied conditions. A save defect can erase work without an obvious warning, while a visual cover-up can leave supposedly hidden information in the file. Meyer’s tests show why these failures may be difficult to spot with a few sample documents: they emerged across a larger set of real PDFs and during review of how redaction worked.

The project also puts limits on what can be concluded from a rapid AI-assisted build. Meyer says the tool became more robust through testing, redesign and repeated review, but he has not measured time or cost savings. Docestra remains private and is not available to download, so readers cannot independently reproduce the results from the account.

Products Worth Considering

Amazon Kindle Scribe Colorsoft 64GB with 11-Inch Color Display and Premium Pen, Graphite

Amazon Kindle Scribe Colorsoft 64GB with 11-Inch Color Display and Premium Pen, Graphite

As an affiliate, we earn on qualifying purchases.

From First Build to Version 1.5.1

Meyer describes “vibe coding” as building software by giving an AI model plain-language instructions and letting it write the code. Docestra began as a personal local app on Nov. 6, with functions including search, markup, page rotation and extraction, combining files, printing, undo and saving. The initial breadth of features did not establish that the tool worked on every kind of document.

According to the development timeline, the 930-file robustness run and save fix came on Nov. 6. The encrypted-file decision followed on Nov. 7, and the redaction redesign took place on Nov. 8. Version 1.5.1 was recorded on Nov. 12. These dates and test counts come from Meyer’s project records, commit history and version plans, rather than an independent audit.

“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 from the project

Independent Checks Remain Limited

The reported test outcomes come from Meyer’s own project log. The source material does not provide the full test corpus, detailed test methods or the independent reviewer’s complete findings, so readers cannot assess how representative the PDFs were or reproduce the reported results. “Zero leaks” refers to the reported test of the redesigned redaction feature across 917 PDFs; it does not establish performance on every document.

It is also unclear how Docestra performs outside Meyer’s local setup, since it is private and unavailable for download. The account gives no measured comparison of development time, costs or productivity gains. Version 2.0 was still in development on Nov. 13, and its planned shared PDF engine for AI agents was not a finished feature.

Version 2.0 Is Still in Development

Meyer says work on Docestra 2.0 began on Nov. 13, 2026, with a plan for one shared PDF engine for the project’s AI agents. The version was unfinished at the time of his account. His project log identifies no release date or public availability plan, so its next milestone remains unclear.

Key Questions

What is Docestra?

It is a private PDF editor that Meyer says he built with AI coding agents for use on his Mac. He says it runs locally in a browser and is not available to download.

What did testing find?

Meyer reports that a run across 930 real PDFs found a save defect that could make annotations disappear. He also found that 52 of 69 encrypted test PDFs prohibited annotation under their authors’ restrictions.

Did the first white-out tool remove sensitive text?

No. Meyer says it covered content visually while leaving the underlying text in the PDF. He later built a separate redaction feature and reports no leaks in a test of 917 PDFs.

Has Docestra 2.0 been released?

No release is reported. Meyer says development began on Nov. 13, 2026, and that version 2.0 was still in progress.

Source: Thorsten Meyer AI

You May Also Like

ChatGPT’s December 11 Voice Shift: What Users Need To Know

OpenAI will remove Advanced and Standard voice from ChatGPT on Dec. 11, 2026, leaving GPT-Live as the remaining mode. Existing chats will stay available.

Best AI-Powered Automation Software Compared

Compare Zapier and Make on AI features, integrations, control, usability, and value to find the better fit for your workflows.

COMMENTARY: Weekend Reads: Meta’s AI Revolt, China’s Power Play, Fed’s Messaging Fight

Analyzing recent trends: reports of AI unrest at Meta, China’s strategic maneuvers, and the Fed’s cautious communication—what’s confirmed and what’s speculation.

IBM Bob

Search interest in IBM Bob is surging, but details about the development remain unconfirmed. Experts analyze potential implications for AI and enterprise tech.