You Don't See the Foundation: Why AIOps Fails Before It Starts

In partnership with

Observability you can act on, whether you build it or lead it, plus the few signals worth your time.

This week: why the agent in the demo looked so clever, and what it was quietly standing on.

No foundation, no AIOps. Garbage in, garbage out.

This week's Signal Drop

You Don't See the Foundation: Why AIOps Fails Before It Starts. Everyone wants the glossy tower. Nobody wants to fund the year underground.

METRICS & MAYHEM

You Don't See the Foundation: Why AI Ops Fails Before It Starts

00:00

Prefer to watch? It is on YouTube.

What you'll take away:

I lived in Singapore for years, watching tower blocks go up in six months. What you never saw was the eighteen months when the site was just a hole in the ground.
AIOps reads your telemetry and reasons over it. That is the whole trick, which means it is only ever as good as what sits underneath.
The foundation is three unglamorous things. Telemetry that means something across the whole estate, tags that say who owns what, traceability you can follow from change to effect.
An agent on a weak foundation does not fix your gaps. It exposes them, then repeats them back to you in a confident voice, at speed.
The year underground is never wasted. Clean telemetry shortens every incident today, agent or no agent. Nobody cuts a ribbon for a foundation. Fund it anyway.

Try this before your next AIOps conversation. Take one closed incident you already understand. Could an agent have explained it using only the telemetry you emit right now? Could it have seen the change that caused it, and known which team owned the thing that broke? That one incident is the core sample of your foundation.

Signal check

Two that came in this week.

Does everyone run the full pipeline from day one? No. Everyone starts piecemeal, and that is fine. Start with the telemetry that shortens your actual incidents, usually logs and traces on two or three critical journeys, and get those tagged and owned before you go wide.
How do you measure SLOs for an agent? Measure it like a junior engineer doing the same job. Not whether it is up, but whether it is right, and how often a human overrules it. Almost nobody is tracking that yet.

Latest on the blog

The Observability Maturity Assessment: Where Does Your Team Actually Stand? The five levels are not a scorecard to feel good about. They are a map of what to fix next. Read it.
Operational Communication for IT Leaders. What to say, to whom, and when, while the incident is still running. Read it.

What you missed last week

The recent posts, a couple of signals, and what was worth the time.

Worth a listen: Dev Interrupted, 2 August, on agent identity in large enterprises. The step-ladder framing lands where this episode does. You do not hand an agent the keys, you move it up a rung at a time.
The AI Daily Brief, 3 August. DeepSeek's V4 Flash scored 50 on the Artificial Analysis index at $0.03 a task, against $0.59 for GLM 5.2. Cheap inference makes the foundation question sharper, because the constraint moves to your data.
Worth a read: The Telemetry Debt Crisis, Cloud Native Now, 5 August. Telemetry debt is the gap between what your system emits and what a human can act on.
EBIT-T-DA: AI's hidden line item, SecurityBrief Australia, 4 August. Tokens as a line on the P&L, with 84% of respondents reporting more than six points of gross-margin erosion.
Dynatrace execs dish details on Bluebox autonomous AI, TechTarget, 29 July. The counterpoint: a major vendor shipping the end state that assumes the foundation is already there.

Tech Tuesday

Back this Tuesday: wide events, and what Observability 2.0 actually means. Are the three pillars dead, or did we just stop asking the question three separate tools cannot answer at 3am? The written anchor goes live with the drop.

That is the week. Next time, what happens when the foundation is sound and the thing standing on it still falls over.

Allan

PS: If an agent cannot explain an incident you already understand, it is not going to explain one you do not.

Everything in one place

New here? Start with the free chapter.

Read the free chapter →

Your employees are connecting AI to everything. Now what?

ChatGPT and Claude don't just answer questions anymore. Employees are connecting them directly to Notion, Linear, Jira, and the rest of your stack. The AI can read, write, and take actions on company data. Most IT and security teams have no visibility into any of it.

Harmonic Security Connectors changes that. It sits inline with every AI-to-app connection, so you see each call, control what data moves, and block destructive actions before they happen. Employees notice nothing different.

See what's actually running across your business in a live demo.

See what's connected