NDEVR · Where we stand
A powerful tool. A dangerous one. Yours to control.
This is where NDEVR stands on artificial intelligence: what we think it is, why we think it’s dangerous, why we build with it anyway, and why every OWL account ships with a real off switch.
The stance, in four parts
Our AI Stance
A tool we must learn to wield.
AI is a powerful tool. We mean both words. Powerful: it collapses years of skilled labor into minutes and puts a tireless, competent generalist at every keyboard. Tool: it has no goals of its own, no mandate, no judgment. It does what the hand holding it points it at. Brilliantly or carelessly, either way.
Powerful tools have never been casual acquaintances; a chainsaw earns a different kind of respect than a screwdriver. So the obligation is neither worship nor prohibition. It is to learn it: build the habits, the guardrails, and the culture that keep the human deciding while the tool executes. A tool must never wield its operator. The moment it does, the operator has already failed.
Its danger is power without buy-in.
Every earlier form of sweeping power needed people. Armies needed soldiers. Factories needed workers. States needed clerks; publishing empires needed writers, printers, and paper routes. Each pair of hands was a quiet veto: power over humanity, at scale, required buy-in from humanity, at scale. It was abused anyway. But it first had to be sold to thousands of participants, and sometimes the thousands said no.
AI is the first lever of sweeping power that removes the hands. A few individuals with enough compute can now persuade, surveil, produce, and decide at a scale that used to take institutions full of consenting people, and almost nobody has to agree along the way. That is the danger as we see it. Not a machine with intentions: a handful of operators freed from the old need for everyone else’s consent.
Before: sweeping power ran on thousands of participants, and every one of them could refuse.
Now: a few operators and their compute. Nobody else has to say yes.
The computer question, again.
Here is the uncomfortable one. We build software for a living, and we cannot cleanly tell you the computer improved the health and wellbeing of the average person. It connected the world and atomized it. It freed us from drudgery and chained us to screens. It put the whole of human knowledge one tap away, and made human attention the most strip-mined resource on earth. Fifty years in, honest people still disagree about the net effect.
We extend the same honesty to AI: it is genuinely questionable whether it improves the life of the average person. Anyone who tells you the answer is obvious, in either direction, is selling something. Holding that doubt while still building is not a contradiction. It’s the job.
Why we build with it anyway.
A tool this powerful does not go away because thoughtful people decline it. Refusal doesn’t retire the tool. It concedes it: the advantages, and the direction, pass to whoever picked it up with the least hesitation. The surest way for careful hands to lose the wheel is to take them off it.
So we engage: deliberately, on our terms. We use AI to build, and we shape what we ship into the kind of AI we’d accept ourselves: pointed at the person using it, never trained on your data, never silently on, off wherever you say, on-device where we can manage it, and honest about the boundary where we can’t.
Not for everyone. On purpose.
We made our choice. We don’t make yours.
Some of you have weighed the same facts and landed somewhere else: no AI near your data, full stop. We built OWL so that answer costs you nothing. QUAIL, our AI assistant, is a guest in your OWL, and guests follow house rules.
Off entirely. One master switch in QUAIL settings turns QUAIL off across OWL: chat, documents, mail, media, all of it. Manual controls stay; nothing else about OWL gets worse.
Off wherever you say. Fifteen per-surface switches. Keep QUAIL in the log viewer and bar it from your mail. Your rooms, your rules.
Nothing before consent. No QUAIL surface reads anything until you explicitly enable it, after being told in plain words exactly what it will receive. It never gets your account identity, and your prompts never train models.
Local where possible. A growing set of routines runs entirely on your device: photo restoration, OCR, background removal, album sorting. No upload, no gateway, no tokens.
The preference stays home. Your QUAIL settings live on your device, not in our database. We don’t even keep a record of your “no.”
And the honest boundary:
Cloud AI is not zero-knowledge. When QUAIL answers, our server relays your request, and only what you enabled, to an AI model outside OWL’s end-to-end encryption boundary. That is the real cost of the feature; it’s why every surface asks first, and why the off switch is real. Details live in our AI policy and our plain-English guide to AI training and your data.
You decide whether and where QUAIL is available. Try it: the switches work.
The master switch. Off means off, everywhere.
Direct QUAIL conversations and Ask QUAIL actions.
Drafting and editing assistance.
Draft and edit email with QUAIL.
Identify media and generate subtitles.
…plus eleven more surfaces, each with its own switch.
A working copy of the real panel in your OWL dashboard. The preference is stored on your device; we never see it.
“I don’t love every fact about this era. But I’d rather have our hands on these tools, pointed at your benefit, than concede them to people who never paused to worry. And if you’d rather keep AI out of your OWL entirely: the switch is real, it’s yours, and we’ll never quietly flip it back.”