Blog

8 May 2026 Architecture

The future of server administration: AI-driven, with human oversight

How AI-driven systems move us from reactive monitoring to proactive, intelligent infrastructure management — without handing over the keys.

For years, system administration has run on a tedious, reactive model. A monitor flags an issue. An alert is sent. A human engineer logs in, hunts through error logs, and manually types out a fix. As server environments grow more complex, this manual troubleshooting cycle stops being merely annoying and starts being unsustainable.

Enter AutOps.

AutOps represents a shift in how infrastructure gets managed. By embedding a large language model directly into the monitoring loop, we move past simple red-light/green-light dashboards. The system continuously analyses health metrics, autonomously fetches standard and error logs, and diagnoses the root cause of an anomaly in seconds rather than in however long it takes somebody to wake up and find their laptop.

But real innovation requires safety. Rather than risking catastrophic outages through fully autonomous, runaway AI changes, AutOps uses a strict human-in-the-loop architecture. This is the part we consider most important, and it is the part that gets the least attention in most AIOps marketing.

When an issue is diagnosed, the AI stages the exact remediation command and sends a proposal directly to your phone via Discord. With a simple !approve reply, the system executes the fix and verifies it worked. The balance is deliberate: automated speed where speed is safe, human judgement where the consequences are irreversible. The practical effect is that engineers can safely handle daily fires from their pockets — and that a wrong diagnosis costs a declined prompt instead of an outage.

The trade-off is honest, too. Approval-gated execution means that if nobody is available to approve, resolution waits. For destructive operations on production infrastructure, we think that is the right side of the trade to be on.

1 May 2026 Engineering

Orchestrating automation: building with n8n and Docker

A look at the core architecture, and why containerisation plus visual workflows beat a pile of shell scripts for infrastructure automation.

Building a reliable virtual systems administrator requires a solid foundation. For AutOps, that foundation is Docker for isolation and n8n for orchestration.

Security and isolation matter more than usual when the software you are writing has SSH access to production. By using Docker, every core component of the AutOps engine — from the custom Python messaging bridge to the AI logic — runs in its own containerised environment. That minimises dependency conflicts, safeguards the state databases holding pending actions, and means a failure in one component cannot cascade into the others or into the host.

To bridge the gap between AI diagnostic reasoning and actual server execution, we chose n8n. Unlike rigid traditional scripting, n8n provides a stateless visual orchestration layer, which lets the system trigger complex multi-step workflows from real-time server data. Background SSH health checks, terminal output parsing, routing AI-generated proposals back to a human — all of it becomes a workflow you can read, audit and modify rather than a script somebody has to reverse-engineer later.

That last property turned out to matter more than we expected. When we hit a protocol mismatch between Discord's stateful gateway and n8n's stateless execution model, being able to see the whole pipeline visually is what made the fix — a dedicated Python bridge daemon translating gateway events into webhooks — obvious rather than mysterious. We wrote that story up in more detail on the architecture page.

The wider lesson: for a system whose whole value proposition is transparency about what it is doing to your servers, choosing an orchestration layer that is itself transparent was not a stylistic preference. It was the same decision, applied one level down.