Cadres IT Operations & Infrastructure
Sheet LOG-01 Rev 2026.08
Start Trial

Entry LOG-01 — · 4 min · Pete

It Started With Patching

Each server patch was costing us 30 to 60 minutes of someone checking boxes. That's where Cadres came from.

foundingvision

If you’ve spent any real time running IT operations, you know nobody notices us until something breaks. We’re the dial tone. When it works, nobody thinks about it. When it doesn’t, everyone has an opinion.

The tools we use haven’t moved much in ten years either. Most of them look like they were designed around 2014 and have been picking up features ever since. The UX comes last, the integrations are held together with tape, and the “automation” is usually a script runner with a GUI on top. When a vendor does ship something new, they mostly bought it.

Thirty to sixty minutes a server

A while back I was shopping for a new patching tool for a large enterprise. What I wanted was patching that ran on its own. What I had was a process that cost 30 to 60 minutes of someone’s time per server: pre-checks, post-checks, comparing before and after, making sure services came back, checking disk, making sure nothing had drifted.

Apart from covering the risk, that time added nothing. Someone followed an SOP line by line, took screenshots, pasted them into the ticket and moved to the next server.

Why can’t that be automated?

I don’t mean automated the way vendors use the word, where someone still has to watch it and click approve at every stage. I mean capture the system state, patch, capture it again and compare. If something looks wrong, roll back. If it’s fine, close the ticket. A person only gets pulled in when there’s a decision to make.

One question led to the next

That was the starting point, and it didn’t stay small for long.

If patching can check its own work, why is someone still hand-triaging alerts that have an obvious fix? Why doesn’t the ticketing system know what monitoring knows? Why does the compliance scanner still flag a control that a patch fixed an hour ago?

The idea I landed on was simple enough to write on a whiteboard: automate IT operations, and bring a human in only by exception.

I’m not trying to replace people. Skilled engineers shouldn’t be doing work that a decent state machine can do, and they should have their time back for the problems that need judgment.

We’re not an AI company

The industry has lost the plot on this one, so I’ll say it once. Cadres isn’t an AI company and we aren’t selling AI. You don’t need it for basic operations automation.

Comparing two system snapshots doesn’t need a large language model. Checking whether a service restarted after a patch doesn’t need machine learning. Following a decision tree your L1 team already wrote down in a runbook doesn’t need anything “intelligent” at all.

It needs a platform that has the data in one place and can run the logic. That’s an engineering problem.

If you do want an AI agent in your workflows (to enrich data, triage alerts or suggest fixes), you can bring your own. The workflow engine supports it. Your agent, your model, your data. We built the plumbing; what you run through it is your call.

Building instead of buying and stitching

Once you decide the answer is one connected platform rather than eight tools glued together with APIs, you have to choose: buy products and stitch them together, or build from scratch.

We built.

It wasn’t the easy option. It’s close to building five or eight products at once. What you get for it is one data model underneath everything. The patching engine knows what a compliance policy is. The service desk knows what a host is. Monitoring knows when a change window is open. Nothing syncs every five minutes and hopes it didn’t miss anything.

So when an alert fires, the system already knows the host’s patch status, compliance state, recent changes and who has privileged access to it. It doesn’t have to call four APIs to find out. It’s all in one database, under one permission model, written to one audit trail.

You can’t get that by buying companies and putting a single sign-on page in front of them.

What’s in it

The platform covers the day-to-day of an IT operations team:

  • Patching with ring-based rollout, canary gates, circuit breakers and automatic rollback, plus the before-and-after state comparison. People get involved when something fails.
  • Monitoring with infrastructure metrics, alert routing, escalation chains and automated response.
  • Service desk with incidents, problems, changes, service requests and SLAs. A proper ITSM, not a ticket list.
  • Privileged access management with time-bound grants, approvals, session recording and an encrypted credential vault.
  • Compliance scanning against CIS and STIG benchmarks, with one-click remediation and audit evidence collected as it runs.
  • Vulnerability tracking tied into patching, so remediation isn’t a separate process.
  • Network discovery for everything on the network: switches, routers and firewalls as well as servers.
  • Workflow automation with approval gates, conditions, parallel steps and PAM built in. This is what makes the human-by-exception model work.
  • Change management with risk assessment, maintenance windows and an emergency change path.

It’s all multi-tenant from day one. Customer isolation has to be enforced properly, not by filtering on account_id and hoping nobody writes a bad query.

What this blog is for

I’ll use this blog to write about how we build things and why, including what we get wrong. Some of it will be engineering, some product decisions, and some opinions about the IT operations industry that it probably won’t enjoy.

I started Cadres because I got tired of waiting for someone else to fix the tooling. The people who run infrastructure have put up with the same tools for a long time.

If that sounds like your world, start a trial. I’d rather build for people who already feel this than spend my time convincing people it’s a problem.