Intelligence artificielle, the French word for AI, literally means “artificially intelligent.” It isn’t itself intelligent, and Benjamin is clear to make that distinction in his DevOps work.
“In French, we say intelligence artificielle, but this word has to be swiped. It’s artificially intelligent. It’s not intelligent itself.”
Benjamin Caouren, a fractional consultant who oversees a DevOps team at a $7.4 billion French retailer with over 500 stores, emphasizes that this phrasing is crucial.
Even as APA becomes more self-directed, it still requires human oversight to ensure that direction leads to strategic outcomes and business goals. That means humans will become increasingly important over time, not less.
He carries this mindset to both managing his team and implementing Application Performance Automation. The future in APA isn’t replacing humans; it’s elevating them from drudgery to doing strategic work that matters.
Just do it right, then do it better
Real APA transformation takes time. Rather than jumping into automating everything, Benjamin’s philosophy is to focus on the parts of your DevOps process that actually need it.
“As long as you know what you are doing, you can do it better,” he says. “And if you have to do it more than once a day, then automate.”
He began applying this approach with his team at Kiabi, initially in non-production environments. They achieved a 60% cost reduction in their non-production environments by utilizing Spot Instances and optimizing resources. Now that they’ve done that, they can move forward with deploying in their production environment in the next few months.
“Everything is code-writing,” says Benjamin. “You are not doing it with a UI. Everything is provided by code.”
The transformation took months, not the 30 minutes most vendors promise, but once configured, the platform runs with zero daily management, exactly what Benjamin means by ”do it right.”
That’s Benjamin’s approach to APA, in his words: “Just do it. Then, do it right. Then, do it better.”
The intelligence that needs direction
Benjamin insists that AI isn’t truly intelligent because it requires human input to work. It needs a direction, a task to be pointed towards, and proper guidance on what the company wants to achieve. More than anything, it needs human context.
Consider Kiabi’s Java applications: when these programs start up, they need ten times the resources they use when running, similar to a car needing extra power to start. After ten seconds, usage drops to 10%. Automation sees low usage and cuts resources, causing crashes at the next startup.
It’s tasks like these where Benjamin holds firm that AI needs human direction. The automation was technically doing what it was supposed to: eliminating unused resources. However, without understanding Java’s initialization pattern, it created an endless crash loop.
Only engineers who actually know how their systems work can explain why that 90% waste is actually necessary. The human insight of DevOps engineers recognized the problem and developed solutions, such as Kubernetes’s new-in-place pod resizing, which allows applications to use different resource levels without requiring a restart.
“It’s more important to be experts than being able to do everything on the platform,” says Benjamin.
The true power of APA, and what became Benjamin’s ‘aha moment’ with Cast AI, is in visibility. The platform revealed what native GKE tools never could: the true cost per workload, namespace by namespace.
“From my point of view, it was the first time I had such information,” Benjamin recalls, describing his ability to drill down from clusters to namespaces to individual workloads.
Before this transparency, teams knew costs were high but not why. With proper visibility, drilling from clusters to namespaces to workloads, Benjamin discovered shocking patterns, such as single namespaces consuming 40% of the resources.
This transforms everything. Instead of Benjamin’s “180% everywhere” scramble, engineers see exactly what’s breaking. That 2amJava crash? They immediately spot the resource limit hit during initialization. Hours become minutes.
More importantly, visibility enables proactive design. “Is it mandatory for me to be connected every day to check if it’s working fine or not?” he asks. When you see patterns, such as Java’s startup spikes and weekend drops, you automate before problems occur. Engineers evolve from firefighters to architects, building self-correcting systems instead of constantly fixing breaks.
Two languages, one goal
The money conversation is what gets buy-in from leadership, but for Benjamin, the real game-changer is using APA to bridge DevOps and FinOps. With the visibility APA affords his team, he can see a direct line from clusters to workloads to costs.
That visibility gives freedom. He can stop constantly monitoring his instances and focus on innovation.
The numbers speak volumes: Spot Instances deliver 60-90% discounts, and smart scheduling through Cast AI’s Rebalancer saves 100 vCPUs for 12 hours during nights and weekends.
After the initial setup, it requires zero daily management. The multi-node pool flexibility even overcame GKE’s typical single-node type limitation, letting Kiabi match resources to actual needs.
When infrastructure costs drop by half AND engineers stop getting paged for predictable issues, the real transformation begins. Time previously spent firefighting now goes to innovation, a virtuous cycle where cost savings multiply with recovered human hours.
Benjamin likens it to using Perplexity instead of ChatGPT or Google; the former cites sources directly, so he can see where the information is coming from. Similarly, APA provides traceability, which fosters trust.
Understanding your tools is more important than just using them.
Preparing for autonomous infrastructure
The future needs more specialists, not generalists.
APA will handle complexity while humans provide the context. Even as APA handles the continual monitoring and optimization, humans will provide it with the environment in which to work and ensure the automations are running properly. From there, the importance of APA will shift from cost reduction to workload optimization.
That oversight and strategic guidance will require APA-ready DevOps engineers who are grounded in the fundamentals, enabling them to use APA with intention.
“I want all my developers to know about the infrastructure, because that’s the foundation,” says Benjamin. “All applications runs with an infrastructure, whether you’re aware of it or not.”
From toil to impact
Benjamin’s APA roadmap for the retailer enables his engineers to stop putting out fires and instead prevent them from happening in the first place.
Instead of having to be “180% everywhere,” he wants his DevOps team to focus on architecture decisions that matter and use APA to help them do it. The shift from reactive to proactive means they can spend less time on maintenance and more time on innovative solutions.
“At the end, we are still human, so we cannot be everywhere,” says Benjamin. But with APA, we don’t need to be. We just need to be in the right places.
The more autonomous the systems become, the more critical human strategy is in the process.
Engineers who understand both infrastructure fundamentals and business context become irreplaceable, not irrelevant. Not because they keep the lights on, but because they decide where the lights should point.
Start where Benjamin’s team did, with visibility. Understand what you’re spending and why. Then apply Benjamin’s framework: “do it, do it right, do it better.” The path from reactive to proactive isn’t a leap; it’s a progression.
Kubernetes cost optimization
Monitor organization-wide and cluster-level resource spending. Automate resource allocation and scale instantly with zero downtime.
