I was recently at a keynote on AI and cloud native applications when the speaker dropped a line that stuck with me: 0 × AI = 0.
The point was simple. Whatever you feed an AI tool — data, context, a half-formed question — gets multiplied by the model’s capability. But multiply anything by zero, and you still get zero. No amount of AI horsepower turns bad input into good output.
For engineers who have spent years worrying that AI is coming for their jobs, the speaker framed this as reassurance: your experience is what makes the input non-zero. The more relevant, accurate, and well-framed the data and questions you bring, the more relevant, accurate, and realistic what comes back.
I think that is true. I also think it is only half the equation.
The equation everyone quotes
GIGO — garbage in, garbage out — is not a new idea. It has been true of every system since the first punch card. What has changed is that AI makes the consequences of bad input feel invisible, because the output still looks polished. A spreadsheet full of wrong numbers looks wrong. An AI-generated summary of wrong numbers looks confident, well-formatted, and entirely plausible — right up until someone acts on it.
So “0 × AI = 0” is a good warning. But in nearly two and a half decades of being in middle of private cloud operations, I have learned that clean data alone is not the multiplier either. You can hand an AI tool a perfectly accurate dataset and still get a useless — or actively misleading — answer, because the AI does not know which numbers in that dataset actually matter.
The multiplier the keynote left out
Here is a scenario I run into constantly in VCF Operations work: reclaiming wasted capacity in VCF environment.
Modern VCF Operations dashboards and custom Super Metrics can surface an enormous amount of useful metrics for this purpose — CPU, memory, storage utilization, contention, oversized VMs, idle workloads, all of it clean, current, and accurate. Feed that raw metrics to an AI tool and ask “where can we reclaim capacity to free up ESX hosts to build a new VCF 9 management domain?” and you will get an answer on how many hosts which can be reclaimed. It might even sound authoritative.
But that answer is only as good as the judgment behind how it is framed. A host showing low average CPU utilization looks like reclaimable capacity — until you know it is carrying a bursty, latency-sensitive workload where headroom is the whole point, not waste. A memory metric that looks “safe” to reclaim might be masking a ballooning problem that only shows up under load. Suggestion of reclaiming one host from a cluster of 3 was not practical as I found. The data was never wrong. The interpretation of what the data means is where experience does the work AI cannot.
That’s the second multiplier missing from the keynote’s version of the equation:
(Data × Domain Judgment) × AI = Insight
Bad interpretation of good data is still a zero. It is just a zero that is harder to spot, because it arrives dressed up as a confident recommendation.
Where this connects to GreenOps
This is not an abstract point for me — it is the whole premise of the capacity reclamation work I build for customers. The goal is straightforward: use VCF Operations dashboards to identify genuinely wasted capacity in an existing environment, so customers can repurpose physical hosts for VCF 9 management domains instead of procuring new hardware.
That is a GreenOps outcome — less new hardware, lower energy and embodied-carbon footprint, better use of what is already running — but it only holds up if the “wasted capacity” identified is actually wasted. An AI tool pointed at raw utilization metrics, without an operator’s understanding of workload behavior, peak patterns, and what headroom is actually protecting, will happily recommend reclaiming capacity that is not safe to reclaim. Get that wrong, and you have not achieved sustainability — you have just moved the risk from “too much hardware” to “not enough.”
The AI accelerates the analysis. It does not replace the judgment about what the analysis means. That judgment is built from years of watching how environments actually behave under real conditions, not just what a dashboard says at a single point in time.
The reassurance, earned rather than asserted
So back to the keynote’s core message for worried engineers: is it right that experience is what makes AI’s output valuable rather than a threat to your relevance?
Yes — but not just because experience helps you provide better input. It is because experience is what lets you judge the output — to look at an AI-generated recommendation and know, immediately, whether it is the kind of thing that would actually hold up in production or the kind of thing that sounds right and is not.
That’s not a skill AI is close to replacing. It’s the skill that decides whether AI is useful at all.
Don’t Fear the Multiplier—Be the Base Value
AI will not replace engineers. But engineers who leverage their experience as the base value in the equation will run circles around those who ignore the shift.
Your years in the trenches—the broken deployments, the race conditions, the network partitions, and the architectural migrations—are not outdated baggage. They are the exact non-zero variable required to make modern tooling produce anything of real value.
Bring the domain intelligence, enforce the constraints, and let the tool handle the mechanical heavy lifting.
0 x AI = 0. Make sure your number is as high as possible.