Selling Software by Subtraction

Two million downloads. No chatbot included.

I review AI tooling for a living. Most weeks that means installing something that promises to rewrite my emails, summarize my meetings, and generate a slide deck from a half-formed thought. So when LibreOffice shipped version 26.8 on August 26, 2026 and listed the absence of generative AI as a headline feature, my first reaction was that it read like a stunt. My second reaction, after watching the download numbers, was that it is the most honest product positioning I have seen all year.

The numbers are not subtle. More than a million downloads in the first week after the announcement. Over two million since. For a free office suite that has existed for years without needing a marketing moment, that is a sharp spike, and the trigger was a feature defined entirely by what it does not do.

What “no AI” actually means here

It is worth being precise, because “no AI” is a phrase that gets stretched. In LibreOffice 26.8, the claim covers three concrete things:

• No generative AI features in the application.
• No transmission of your documents to remote servers.
• No network access required for the software to function.

That third point is the one I would underline. Plenty of tools say they respect your privacy. Far fewer are architecturally incapable of phoning home, because they do not need a connection to work at all. That is a verifiable property, not a policy promise. Policies change when a funding round closes. Software that runs fine with the Wi-Fi off keeps running fine with the Wi-Fi off.

The Document Foundation has also been careful not to plant a flag it cannot defend. Its position is not that AI is inherently bad. It is that office software used by tens of millions of people carries obligations, and certain principles have to be satisfied before AI gets added to that kind of product. That is a narrower and more defensible claim than blanket refusal, and it is the part most of the coverage skipped past.

Why this landed so hard

In my inbox, the most common complaint about AI features in productivity tools is not that they are inaccurate. It is that nobody asked. The assistant panel appears after an update. The sidebar you did not open starts suggesting completions. A dialog asks whether you want to enable smart suggestions, and the only honest answer is that you have no idea what that costs you in terms of where your text goes.

Office documents are a bad place to be uncertain. People put contracts, medical notes, salary spreadsheets, legal filings, and unpublished manuscripts in word processors. The question “does this upload my file somewhere” is not paranoia in that context, it is basic diligence. For a lot of users, and especially for people buying software on behalf of organizations with compliance requirements, the ability to answer that question with a flat no has real value.

So LibreOffice did something unusual. It stopped treating that answer as a technical footnote and started treating it as a product claim. Two million downloads suggest the demand was sitting there unserved.

The reviewer’s read

I do not think this is an anti-AI story, and I would push back on anyone framing it that way. I use language models constantly. They are genuinely good at some of the work they are sold for. What this episode exposes is a gap between how vendors think about AI features and how users experience them.

Vendors think in terms of capability added. Users experience it as surface area added: more code paths, more data flows, more things to audit, more defaults to check after every update. When the capability is useful, that trade is fine. When it is a suggestion engine nobody requested, the trade is bad, and users notice.

The practical lesson for anyone shipping a tool is that “AI-free” is now a positioning option with proven demand behind it. Not for every product, and not for every audience. But the assumption that adding AI features is always net positive for adoption just took a public hit, and the counterexample came from a free office suite with no advertising budget.

What I would watch next

Whether this holds depends on things we cannot measure yet. Download spikes measure curiosity and goodwill, not retention. A million people trying something because they liked its principles is different from a million people staying because the software does their work well. Typography improvements and wider support for the world’s writing systems shipped in the same release, which matters more for daily use than any announcement.

Still, for a release defined by restraint, the response says something worth hearing. Sometimes the best feature you can ship is the one you decided not to build.

🕒 Published: October 7, 2026