Juan Pablo Olivera - Cloudacio Insights

Illustration: A row of identical empty crates on a conveyor belt, and beside it one large open crate holding many small nested compartments, most of them sealed shut and a few standing open

Cloud

Why Kubernetes keeps killing your AI agents

You can run AI agents on Kubernetes, with conditions. The pod-per-agent assumption is what breaks first, and for most teams the fix is the signal their autoscaler reads, not a new substrate.

Illustration: A stopwatch mid-sweep beside a balance scale whose two pans rest perfectly level, set on a surface of stacked cluster nodes

AI

What a Kubernetes AI benchmark measured, and what it did not

Read-only cluster context cut mean time to a Kubernetes diagnosis from 191 seconds to 64. Correct diagnoses went from 45 of 50 to 46. What you buy with that speed, and how to measure your own agent before you sign for a better one.

Illustration: Two separate paper invoices lying side by side, each threaded onto the same single pipe that runs through both of them and off the edge of the frame

AI

Why one AI question shows up on two different bills

The survey everyone quotes on surprise AI costs has moved, and the stale version is still circulating. The useful part is the join: the system tables and label placements that trace one agent task across both the model bill and the warehouse bill on Snowflake, Databricks and BigQuery.

Illustration: a set of mechanical gauges mounted on a control panel, most of their dials blank and unmarked, one gauge showing a clearly calibrated scale

Engineering

Your Kubernetes agent's revert rate is the metric that matters

The architecture for autonomous Kubernetes remediation is settled: the model diagnoses, GitOps acts, RBAC bounds. What nobody can tell you is how often their agent is wrong. Here are the metrics that can embarrass you, and why fixing your resource requests may delete more toil than the agent does.

Illustration: A continuous pipe carrying liquid from a row of metered gauges into open air, where it breaks into separate unmeasured streams falling past the last gauge

AI

Why AI cost forecasts miss, and what to measure instead

98% of enterprises track AI cost and 11% can forecast it within 10%. The coverage data in the same survey says that is an instrumentation gap, not a modelling one, and a tighter forecast is the wrong thing to want.

Illustration: a long row of identical unmarked doors, every one of them being tried by a mechanical arm, with one door at the end standing open and forgotten

Cloud

Attacking you costs $5 per try

An operator pointed AI agents at more than a thousand e-commerce sites for an estimated $5 a try and about 23 unattended hours per win. The control that changed in value is knowing what you expose.

Terminal: aws ec2 run-instances failing with InsufficientInstanceCapacity

Cloud

Why your EC2 quota doesn't guarantee capacity

AWS capacity is not sold out, but the slack is gone, and that makes region and instance family design decisions again. What the error codes, the SLAs and the commitment instruments actually say.

Illustration: Three nested gates of decreasing width standing in a row along a single track, with a small sealed key resting on the ground just beyond the last gate

AI

Give every AI agent an owner before it gets a credential

Surveys keep finding the same gap: nearly everyone is confident their agent controls are sound, and about a third can show it. Three decisions close most of that gap, and all three are cheaper before the credential exists than after.

Illustration: a neat stack of building blocks where most blocks are hollow outlines and only a few at the base are solid

Cloud

How to read a cloud bill that jumped and find out why

A 40% jump rarely has one culprit. How to break the total into places to look, in an order that keeps the product safe.

Illustration: a row of jugs being filled from a single pipe

Engineering

What breaks first when a web app grows from 100 to 10,000 users

Growth tends to fail at fixed limits rather than gradually. Here are four limits a growing product hits, how to spot each one with tools you already have, and why the database connection limit should come first.

Illustration: a single torch beam reaching into a tall wall of drawers, lighting only the one drawer it needs

Why your RAG demo works and your RAG product does not

A RAG demo is tested on questions and documents its builder chose. What changes in production, and why a fixed set of fifty real questions is the first thing to build.

Illustration: a sealed envelope resting on a plinth beside an hourglass whose sand has only just begun to fall

Cloud

How to answer a SOC 2 request before you have a report

What a prospect means when the security questionnaire asks for SOC 2, how Type I and Type II differ in evidence and time, and what you can honestly send while the report is months away.