Should an AI be allowed to fix your own computer?

The dream of IT support is simple: something breaks and it quietly gets fixed, ideally before you've even noticed. AI finally makes that plausible — an assistant that can look at a misbehaving PC, reason about what's wrong, and put it right. The demos are genuinely impressive.

The uncomfortable question sits just underneath the demo, and it's the one that actually matters: how much should you let it do on its own?

Because the same capability that fixes a stuck print spooler can, pointed slightly wrong, disable the thing keeping a laptop secure. "Autonomous" is a selling word; on someone's work machine it's also a risk word. Having built an autonomous support assistant ourselves, here's the line we drew — and why we think anyone deploying one should draw it too.

Consent is not a checkbox you buried

An AI touching a user's machine should ask before it acts, in plain language, and the user should be able to say no. "Approve once, we'll do anything forever" is how trust evaporates the first time something surprising happens. The moment of consent is also the moment the user learns what's about to change — that's a feature, not friction.

Read freely; write carefully

Diagnosing a problem is low-risk: read logs, check settings, measure. Changing things is where the blast radius lives. The safe pattern is a hard split — the AI can investigate on its own, but any change that's hard to undo waits for a yes. Reversible tweaks can flow; irreversible ones stop and ask.

Some things it must never do

• Never ask for passwords, OTPs or banking details — a legitimate support tool has no reason to.
• Never disable security controls to "make something work."
• Never wander off-topic — an IT assistant answers IT questions, not whatever it's prompted into.

Everything is logged, and a human is one step away

Every action the assistant takes should be written down in language a person can audit later, and there must always be a clean handoff to a human when the problem is out of scope or the user simply wants one. Automation that can't explain itself, or can't step aside, isn't ready for someone's livelihood.

Done this way, autonomous support stops being scary and starts being what it should be: the boring, reliable colleague who fixes the small stuff instantly and knows exactly when to call for help.

This is the thinking behind Resolve, our AI IT-support assistant — consent-first, read-before-write, guard-railed, and always escalating to a human when it should.