Fractional AI & Cloud Platform Architect. AI agent platforms, MLOps, Kubernetes, and cloud-native delivery by Charlie Holland. Senior expertise, no permanent hire.
What a company really spends its money on is the gathering: getting everyone pointed the same way, getting a decision to actually happen, getting one change past four teams and out of the door. More work has always meant more people, and more people has always meant more gathering. A working sheepdog doesn't obey that arithmetic. One dog and one handler move several hundred sheep, and another…
Every year, roughly 2% of the whisky in every cask in Scotland breathes out through the wood and is gone. Nobody fights it. The angels' share, they call it, the price of letting the stuff mature at all. The economy runs the same trick on productivity, except our angels get greedier every year. Which is why Keynes promised your grandparents a fifteen-hour week and you're reading this between two…
There's a lump of granite off the Ayrshire coast that used to be a volcano. Sixty million years of weather wore it down to the one part hard enough to matter, and we've made the world's curling stones out of it ever since. Software does the opposite. Nothing ever erodes, every layer you've ever shipped is still down there needing somebody to mind it, and you pay for all of it in headcount and…
Every enterprise technology programme for forty years has been the same project: getting what people know out of their heads and into a system. AI is just the first tool sharp enough to reach the knowledge we always called too skilled to automate. Here's what the rising water does when it finally reaches the high ground, and why I help build the thing that worries me.
The board has seen the demo and wants the autonomous agent live by Q3. That is not how any of this works. Here is the map I actually use to deliver AI: where it goes, and how to reach it without becoming one of the 40% of projects that get cancelled.
You size a GKE cluster for thirty small stateful workloads, each with a modest PVC. You expect them to pack onto the nodes neatly. Instead half the nodes are idling, a quarter of your pods are Pending, and the error is about 'volume attach limits.' The number, when you dig into it, is fifteen. Why fifteen? Good question.
Eighteen months ago, a senior exec at a global consulting firm asked me whether we could build RAGs. His customers were banging down the doors. The question is quietly awkward now — and most people still asking it don't actually need what they think they're asking for.
You put your AI agent in a container. You put the container in Kubernetes. You added network policies and RBAC. Then an exec asked it to build a sentiment analysis model and it escalated its way through your entire cloud account. Here's exactly how.
Agentic AI on your laptop is one thing — worst case it trashes your machine. But enterprises want it in the cloud, at scale, pointed at everything. The problem isn't the power. It's the containment. And this thing is smart enough to read the blueprints of its own cage.
Somewhere between 'ask Claude for a quick opinion' and 'Claude is writing our Jira tickets,' we lost the plot. AI agents are brilliant implementers. They're also confidently wrong about every decision that matters. And when it all falls over, they won't be the ones carrying the bag.
AI agents will get you to dev-done at record speed — right up until you realise you're miles past the actual problem. 80% of projects still spend 80% of their time at 80% complete. Agents don't fix that. They just help you reach it sooner.
Your exec promised a customer that AI would 'just know' the answer to any question across every database in the org. It won't. AI is a model of reality — and there's always a gap. The question is whether you know what it is.
Too many architecture reviews are just debates about which vendor's service logo to put on the diagram. The real value of an architect is knowing when the menu is the wrong starting point.
We never really needed ten thousand DevOps engineers. The market's not dying — it's clearing out the noise. When the dust settles, the builders will still be here.
Google says Vertex AI is the managed future of ML on GCP. They're probably right. But the migration from self-hosted KubeFlow is 6-12 months of real work, not the 2-sprint effort that the consultants estimated.
The Highest Paid Person's Opinion isn't always wrong — but it needs a reality check from the people who actually build things. The best organisations make space for 'cool idea, but here's how that actually works.'
VPC Service Controls prevent data exfiltration from GCP projects. Simple concept. Brutal consequences. Everything breaks the first time, and the error messages won't tell you why.
Understand when to use Kubernetes and when to go serverless in your cloud-native architecture.
Tiny Jira tickets. Daily check-ins. Hand-holding. Scrum built the perfect assembly line — and AI loves assembly lines. The question isn't whether AI can do the tickets. It's whether your team is more than the sum of its tickets.
The 'missionaries vs mercenaries' framing is lazy and wrong. After 30 years of freelancing, I've seen full-timers who don't care and contractors who bleed for the outcome. It's not about employment status — it's about who gives a damn.
A thousand tiny Lambda functions scattered everywhere with no ownership, no contracts, and no way to test them locally. Sound familiar? Here's when serverless works, when it doesn't, and the discipline most teams skip.
Pipeline jockeys cranking YAML. Firefighting teams on endless rota. Developers throwing features over the wall. Call it SRE, call it Platform Engineering — if nobody owns production, it's just ops in a frock.
AI-enabled revenue feels a lot like tulip bulbs in 1637. Everyone's hoarding data by the petabyte, hoping it'll bloom into gold. But most of it's weeds — and the only ones smiling are the lads selling shovels.
Most Kubernetes abstractions add complexity for complexity's sake. Operators are the exception — they encode operational knowledge into software and turn 'file a ticket and wait three days' into 'apply a YAML file and get a database in two minutes.'
Most contracts last three to six months. I spent five years as a part-time architect at a century-old industrial manufacturer. Here's what long-term fractional engagement really looks like — the advantages, the risks, and why it's the antithesis of mercenary behaviour.
Post-merger tech consolidation is one of the most expensive, least glamorous problems in enterprise IT. Nobody talks about it because it's not sexy. But it's where careers go to die and budgets go to evaporate.
Everyone preaches 'find vulnerabilities earlier.' The concept is sound. The execution is usually a disaster — every build fails on day one, exception lists grow longer than vulnerability lists, and security becomes a rubber stamp that makes everyone feel safe without actually being safe.
Azure DevOps Pipelines is the CI/CD system enterprise teams end up with because they're an Azure shop. Not because they evaluated the options. Not because they love it. Because it was already there.
When a century-old manufacturer tries to 'go digital,' the cultural collision is spectacular. Manufacturing thinks in years and certifications. Software thinks in sprints and experiments. The smart money respects both.
Most Kubernetes comparisons are feature matrices written by people who've read the docs. This is what the three major managed Kubernetes services actually feel like when you're running production workloads, handling upgrades at 2am, and arguing with IAM policies.
After a year of running KubeFlow across multiple client deployments, I can confidently say: the platform that promises end-to-end ML on Kubernetes delivers end-to-end frustration instead. Here's the post-mortem.
The Infrastructure as Code wars are over. Terraform won. And yet enterprise teams are still clinging to vendor-native tooling like ARM templates and CloudFormation because 'it's supported by the vendor.' Supported doesn't mean good.
Vendor consultants optimise for showcasing their platform's capabilities, not for your team's ability to maintain it. They build reference architectures, not products. The demo passes. Production doesn't.
Not everyone can 'just use the cloud.' Healthcare data sovereignty, air-gapped networks, public sector paranoia — sometimes you're deploying ML inference on a rack server in a basement. Here's what that actually looks like with MicroK8s.
Every organisation wanted an MLOps platform circa 2020. Most had data scientists running Jupyter notebooks on their laptops and calling it a pipeline. Building an airport for one flight a week.
Nobody is running workloads portably across clouds. They're running different things on different clouds and calling it a strategy. Multi-cloud is what happens when nobody's in charge.
I've deployed Kubernetes manifests with both Helm and Kustomize across half a dozen enterprise clients. The debate is tiresome. They solve different problems. Here's when each one actually makes sense.
ECK just went GA and we're running Elasticsearch on Kubernetes in production. It works. It's also the most operationally demanding thing in the entire cluster, and we spend more time babysitting it than everything else combined.
Everyone adopted Istio because everyone else was adopting Istio. Complex, resource-hungry, and solving problems most teams didn't have — it was the poster child for resume-driven development.
Service Fabric got a lot right — stateful services, the actor model, rolling upgrades with health checks. But it lost the ecosystem war to Kubernetes, and being technically superior wasn't enough to save it.
We were using FluxCD to reconcile Kubernetes state from git before anyone called it GitOps. The tooling was raw, the wins were real, and the company that coined the term went bankrupt. Make of that what you will.
The Databricks sales demo runs on a public workspace with default networking and an unlimited budget. Your enterprise Azure environment has none of those things. Here's what the gap actually looks like.
The cloud wars are raging. The secret to coming out unscathed is to back open technologies, think carefully about IaaS vs PaaS, and not get locked in while the giants fight it out.
Since the 1950s we've been looking for better ways to build software. Iterative development, continuous integration, and SaaS delivery have fundamentally changed the feedback loop — and the winners are those who embrace it.
Software as a service is a quiet revolution, dramatically changing the way we do things. The winners accept that they're carrying the can for quality of service — and that's their competitive advantage.