LibreOffice breaks download records after declaring it has no AI features
TL;DR Highlight
LibreOffice 26.8 touted having no AI and passed 1 million downloads in a week
Who Should Read
A product developer deciding whether to add generative AI features to an internal document tool
Core Mechanics
- 26.8 is the most popular update ever
- LibreOffice 26.8, released August 26, had its installer downloaded over 1 million times in a week.
- Updates via Linux distribution repositories are not counted.
- The author bets on 'no AI' as the driver
- The author puts his money on the no-generative-AI statement rather than writing and typography improvements.
- He ties it to the privacy concerns of generative AI.
- TDF says it does not reject AI
- In 'Yes, no AI is now a feature', The Document Foundation (TDF) says it does not reject AI out of hand.
- It says AI does not yet meet the principles for default inclusion.
- Five conditions for built-in AI
- Users control execution, and no content leaves the computer without authorization.
- No telemetry of any kind, and no dependency on a single vendor.
- No compromise on file format, and it must be entirely optional.
Community Reactions
- Opinion that records are meaningless
- Commenters said weekly downloads keep growing with popularity, so calling each high a record means little.
- One said the official stats' monthly view is mostly flat since 2019.
- Opinion that AI agents drove installs
- Several comments said Codex installs bundle LibreOffice and that Codex and web Claude use it to edit docs.
- Another noted OpenAI models use it as a headless editor, which could affect metrics.
- Users' view on AI features
- One said if it can be controlled agentically, it already has the AI features they want.
- Another wanted only sync with a disk file so Claude can work on the XLSX, with no built-in AI.
- Other causes suggested
- Some pointed to Microsoft's offline Office license changes or people switching to Linux.
- One shared installing LibreOffice on their parents' PC after moving it to Linux Mint.
How to Apply
- Audit one AI feature against the 5 conditions today
- Pick one AI feature in your product and check it against TDF's five conditions.
- Note which violations can be turned off in settings.
- Add a user off switch for AI features
- Use a feature flag to disable AI by default or entirely.
- Verify with network logs that no document leaves when disabled.
- Open AI integration via file sync or plugins
- Offer file-based integration or a plugin interface so agents can edit files instead of built-in AI.
- Separate agent traffic from usage metrics
- Split headless calls and automated installs from user counts.
Try It
# Paste into Claude or ChatGPT (a simple checker based on the article's 5 conditions for built-in AI)
Audit my product's AI feature against these 5 conditions.
1. Does the user control execution?
2. Does no content leave the device without authorization?
3. Is there no telemetry?
4. Is there no dependency on a single vendor?
5. Is it fully optional with no file format compromise?
[AI feature description]
{{AI feature description}}
Give a table with pass/fail, a one-line reason per condition, and a fix for each failure.
Next step: pick the easiest failing item and ask for it to be turned into an implementation ticket.Terminology
Related Papers
The Judge Can Debias Itself: Awareness and Self-Debiasing for Reliable LLM-as-a-Judge
Warning an LLM judge about bias or letting it self-critique improved 20 of 21 model–bias combinations.
90% of US working hours are spent on tasks not covered well by AI benchmarks
Mapping 29 benchmarks to job tasks shows 90% of US working hours lack good coverage
Postgres 19 delayed after AI reviewers find more bugs than humans can fix
LLM code review surfaced so many bugs that Postgres 19 slipped to November and 8-10 features were reverted
Show HN: Gemini 4 Argon plays StarCraft
LLMs wrote StarCraft bots in one hour; GPT-6 Astra scored highest.
A Plausible Explanation Is Enough to Mislead Multimodal Judges
Adding plausible explanations to fixed answers cut MLLM judges' wrong-answer rejection by 12.7–13.8 points.
Neural networks are clean
Looking at neurons directly shows most have one main job, challenging superposition.