I’ve spent more than twenty years designing, building and looking after networks, from small offices to some very demanding sites, including work at The Shard, Chelsea FC and Silverstone. Large or small, the pattern was always the same: the genuinely hard problems were rare, and the week disappeared into routine work that had to be done right every single time.
When AI became good enough to reason about real systems, I started asking a simple question: could it do that routine work the way a careful engineer does? Check first. Change only what it’s been allowed to. Prove what it did. Guardian is my answer: twenty years of engineering habits, written down as rules and playbooks an AI has to follow.
It’s also why Guardian is so strict. I’ve seen what a rushed change does to a live network. So it has to diagnose before it touches anything, take a snapshot first, roll back if a check fails, and earn every job by passing an exam built from real incidents.
Bringing those two worlds together is the whole point: an engineer’s judgement, with an AI’s patience to apply it around the clock.
Every task, done by hand first.
Everything Guardian does, from patch nights and lockout loops to DHCP scopes and VPN attacks, has been done by hand on real networks first. Guardian turns that experience into playbooks an AI can follow safely, under a manager’s control.
The safety comes first: permissions decided by fixed rules, hard limits nothing can unlock, a tamper-evident record of every action, and an exam the AI must pass before it touches a client network.
Honest by default.
- Results published in full every run, good and bad, so you can judge for yourself
- Your data stays yours on your premises, exported to you if you leave
- Humans stay in charge always a named manager with the final say