For the last twelve years, my work has been enterprise infrastructure — virtualization platforms, hyperconverged clusters, storage, disaster recovery, identity systems — for banks, government ministries, and large enterprises across Kuwait and the Gulf. Generative AI and agentic AI are now part of that same conversation, and increasingly, the infrastructure side of it is where the real decisions get made.
The AI conversation is missing a seat at the table
Most of what gets written about generative AI and agentic AI comes from one of two directions: data scientists explaining algorithms, or software engineers explaining frameworks. Both are useful. Neither one answers the question that comes up in almost every real deployment conversation with enterprise clients:
"Okay, but where does this actually run, who secures it, and what happens to our data?"
That's an infrastructure question, not a machine learning question. And it's usually the question nobody in the room is fully equipped to answer — because the people building the AI layer often haven't spent years thinking about GPU capacity planning, data residency law, or what a compliance audit actually expects from a production system.
Why this matters more in regulated markets
In banking, government, and healthcare — the sectors this consultancy has spent most of its work in — "just call the API" is rarely the whole answer. There are questions about:
- Where the data physically sits, and whether it can leave the country or the network at all
- How an AI-assisted decision gets logged and audited the same way a human decision would be
- What happens when the model is wrong, and who is accountable for that
- Whether an on-premises deployment is required, and if so, what that actually costs in hardware and operational overhead
These aren't edge cases in the Gulf. They're often the starting requirement.
Where this series is going
I approach generative AI and agentic AI — large language models, retrieval-augmented generation, multi-agent systems, and the tooling ecosystem around them — the same way I approach any new platform I bring into an enterprise environment: understand the concepts properly, then map them onto the infrastructure decisions that make them deployable, secure, and auditable in the environments I actually work in.
This series shares that process, written for other infrastructure and platform people — not to turn you into a data scientist, but to give you the grounding to lead the room when your organization starts asking "should we deploy this, and how?" We'll go roughly in the order the concepts themselves built up over time:
- Rule-based automation, and why it's not actually obsolete
- Machine learning, explained through infrastructure analogies
- Deep learning, and the hardware questions it raises early
- Natural language processing as a translation layer
- What generative AI actually changes architecturally
- Agentic AI — software that acts, not just responds
- Retrieval-augmented generation and vector databases, which look a lot like infrastructure problems in disguise
- Where to actually run all of this — on-prem, cloud, or a hybrid model — especially for regulated industries
- What to evaluate before a bank or ministry greenlights its first AI pilot
If you've spent your career on the infrastructure side and are trying to figure out where you fit into the AI conversation, this series is for you.
Evaluating an AI pilot and need the infrastructure side figured out — data residency, GPU sizing, or secure deployment on OpenShift, VMware, Nutanix, or Google Cloud, Azure, and AWS? Get in touch.