And why the old way of thinking about infrastructure no longer works.
Let me tell you a story about Mia.
She lived with her toy poodle in a seaside town and made incredible ice cream at home.
Small batches, rich, velvety, and deep flavors. The kind that makes stress melt away and makes you feel a little lighter.
She had a big dream. She wanted to make the best ice cream and serve it to Martian people.
But she has one big problem: she was stuck in her home kitchen on Earth.
Martian people were millions of miles away.
And there is no way to make enough ice cream to serve a whole planet.
So Mia asked herself one question:
How do I make the best ice cream possible at scale, away from my home kitchen?
That question has many answers. But finding the right answer alone is worth exploring.
Era 1: The Martian Factory (On-Premise)
First, Mia decided she needed to build a factory on Mars.
She shipped a giant ice cream churner and a massive freezer across a vast cosmic void.
But once Mia set foot on Mars with her toy poodle, she rented an empty building, hired a team to assemble everything, and transformed it into a factory.
It took years, cost a fortune, and nearly bankrupted her.
But eventually, she made it. She had a factory on Mars and could make ice cream to serve Martians.
The Problem:
She had a factory, but many of the same problems on Earth still exist on Mars.
During Martian summer, she couldn’t churn fast enough and would miss out on valuable profits.
During Martian winter, her churner sat idle while the factory still cost money to upkeep.
Whenever the churner broke down, the line was halted, and all hands were on deck to fix it.
The factory on Mars is just like a supersized version of her home kitchen, but equally rigid and restrictive.
Era 2: The Shared Facility (Public Cloud)
One day, Mia realised she didn’t need to maintain a factory or own the machines herself.
Luckily for Mia, she could transfer the factory space’s lease and sell all the machines to someone else for a reasonable price.
She signed a new contract with a shared food manufacturing facility on Mars.
She paid only for the time she used the ice cream churner, the freezer space she used, and the packaging she consumed.
She could quickly ramp up production during Martian summer and scale down during Martin winter.
Her team stopped worrying about broken churners or insufficient freezer space, as the shared facility would handle them.
The Problem:
While there was a bit more flexibility to scale up and down, Mia still couldn’t fully understand and adapt quickly enough to what Martian people love across seasons and special occasions.
The majority of her team’s time was still spent on ensuring the ice cream was well made, affordable.
But very little time could be carved out to build relationships with Martian distributors and customers and develop new flavors that could make Mia’s ice cream popular throughout the planet.
Era 3: The Robot that Helps Mia Think Martian (AI)
Mia heard about a new kind of robot.
Not a single machine that does end-to-end ice cream churning.
But a swarm of robots that could handle the entire production workflow.
One robot could receive milk, eggs, sugar, and fruit from the supplier, while another monitored and adjusted the churning process based on the milk's fat content or the day's humidity.
Other robots handled packaging and move the finished products into the freezer space Mia rents.
Specialized robots could also learn from and analyze data on Martian taste preferences, predicting demand patterns and suggesting new flavor combinations based on what Martians like and dislike.
The robots operate at the shared facility, and Mia paid for the robots based on what each robot does, either by the amount of ice cream produced or the intelligence it brought to her team.
With the intelligence the robots brought, Mia’s team had time to focus on what mattered most:
building relationships with Martian distributors and customers
designing new flavors
improving their services
developing new market-penetration strategies to expand their business across Mars
So What Had Changed?
The New Cost Structure
Era 1 (On-Premise)
Mia’s costs were fixed and predictable.
She bought a giant ice cream churner (capital expenditure a.k.a CAPEX).
She hired team members to run it (labour). Then she paid for the freezer’s electricity (operational expenditure a.k.a. OPEX).
Each area had its own cost, its own budget and its own way of measuring value.
It doesn’t matter how much sales Mia made, she was still paying a lot for the machines.
Era 2 (Public Cloud)
Mia stopped buying and maintaining machines. She started renting them by the hour.
She only paid for churning hours, freezer space and packing. But only for what she consumed.
Her costs became variable and more in-tune with her sales. She could scale up and down easily.
She stopped paying for machines. She was paying for usage.
Era 3 (AI)
The raw materials stay the same, she still pays for the milk, fruit and eggs to make ice cream.
The recipe development process hasn’t changed either.
But the cost of production has completely flipped. She no longer pays for churning hours or freezer space.
She now pays for what each of the robot does and the insight it brings to every scoop.
In the AI world, “tokens” are the new core billing unit and it defines what Mia has to pay to make the robots work for her.
To be clear: the churner, freezer, and packing line still exist and is required to produce ice cream. But they’re no longer the thing Mia thinks about or pays for separately.
She just paid for whatever she asked the robots to do for her business.
And it truly means “whatever”: the good, the bad, the ugly.
The New Superpowers
Era 1 (On-premise)
Back when Mia still had her own factory, her team spent most of their time keeping the factory up and running.
Their days were bogged down with endless tasks ranging from fixing churners, managing inventory, troubleshooting freezer breakdowns.
They were firefighters, not flavour innovators.
Their tools were physical machines that constantly required attention, leaving no time to focus on customers and market strategies.
Era 2 (Public Cloud)
Finally, after moving to the shared facility, her team stopped worrying about broken machines because they could call someone else to deal with it.
They could focus on optimising production and scale efficiently.
They had more time to think about new flavours that Martian would like.
But they were still doing most of the grunt work themselves, just more systematically and efficiently.
Their tools were virtual and largely self-sufficient, but their minds were still buried deep in operations and how to keep things running like a well-oiled machine.
Era 3 (AI)
Mia’s team isn’t swamped by the repetitive work anymore.
They had a swarm of robots that handle the churning, freezing and packaging.
The swarm of robots just handled the repetitive, time-consuming work while providing them the intel they needed to make better decisions.
Overtime, her team has been freed to focus on what matters most to the business.
The robots didn’t replace Mia’s team. They just gave her team superpowers.
In the wider world, we’re seeing the same shift.
Salesforce launched Agentforce to handle automated lead qualification, schedule discovery calls and provide dynamic quoting and pricing for Sales teams.
Microsoft’s Copilot just works in the background to facilitate meetings, draft emails, summarise information scattered throughout chats and emails to assist office workers with certain tasks.
Anthropic launched Claude Code to act like a virtual teammate to independently write, test, debug and execute code for developers.
So many different tools, different use cases, but the pattern is consistent: These are tools to augment human capability, not replace it.
In my own work, I’ve seen this shift play out in real time.
As more agents and AI tools are incorporated into the workflow, the conversation has moved from “Will AI replace my job?” to “How will my role evolve with AI?”
AI is the new superpower. And with a new superpower come new challenges.
These new challenges require a different mindset.
The New Mindset
Era 1 (On-Premise)
Mia was a factory manager.
Her job was to keep the lights on, fix things when they broke, and make sure the giant ice cream churner keeps churning.
Her mindset is operational.
Era 2 (Public Cloud)
Mia became a capacity planner.
Her job was to forecast demand, optimise machine usage, and make sure she isn’t overpaying for idle resources.
Her mindset was financial.
Era 3 (AI)
Mia became a strategist.
Her job was to understand what Martians like and dislike in their ice cream, design new flavors, improve services, and build long-lasting relationships.
She stopped worrying about the factory and started worrying about what her company brings to the market.
Her mindset shifted from “keeping things running with her bill in check” to “creating value.”
A new cost structure. New tools. A new mindset.
It’s a lot to adapt to. But it starts with three simple things.
Three Things Mia Had to Change
From After-the-fact to Real-time Cost Governance
Old Way
Mia knew exactly what things cost. Milk cost X. Churning cost Y. Freezing cost Z.
After years of running the business, she could predict her monthly bill with her eyes closed.
What Needs to Change
With AI, costs are dynamic.
Token-based billing is metered, not fixed, and it can be unpredictable.
One query might be cheap. Another might be expensive. You don’t know until it happens.
It’s the difference between buying a 24-hour Tokyo subway ticket and using a prepaid Suica card on your first day in Tokyo.
The 24-hour ticket gives you unlimited rides for 24 hours from first use for a fixed cost.
The prepaid Suica card is metered. Your balance gets deducted a different amount per ride, depending on distance, time of day, and which lines you used.
And you can’t see your ending balance until you tap out at the gate.
Now imagine it’s not just you who is tapping out. Imagine a fleet of robots, each one tapping in and out hundreds of times a day.
And on the 30th of the month, you might weep looking at the ending balance.
What This Means for You
You need real-time visibility into your costs or a monitoring mechanism that flags when you’re approaching your limit.
You can’t wait until the end of the month to find out you’ve way overspent.
You need to know as it happens so that you can stop it before the cost goes out of control.
From Logs to Reasoning Traceability
Old Way
Mia kept logs, CCTV recordings and ice cream samples.
If something went wrong, she’d check the logs, or the recordings, or the samples to see what happened.
What Needs to Change
With AI, you don’t just need to know what happened. You need to know why the AI made a decision.
That’s “reasoning traceability”. It’s the difference between seeing a robot slow down the churning speed and understanding why it did that.
Was it adjusting for the milk’s high-fat content? Or because the robot was malfunctioning?
Here’s why it matters: You can’t defend a decision you can’t explain.
If a Martian customer asks why their favourite flavour was discontinued, “the robot decided” isn’t an answer.
If a distributor asked why their order was delayed, “the system made a call” isn’t an answer.
If the Martian regulators asks for proof of how you prevent cross-contact between nut-based flavours and nut-free options, “we don’t know, it’s just how the robot does it” isn’t an answer.
“We don’t know” is the worst possible answer, especially in regulated industries. It’s the difference between a compliance officer nodding or escalating their concerns.
What This Means for You
You need to be able to trace the AI’s reasoning path. Not just the output, but also the why behind it.
This is critical for compliance, auditing, and trust. But it’s also critical for something simpler: being able to defend your decisions.
If you can’t explain why your AI did what it did, you can’t stand behind it. And if you can’t stand behind it, you shouldn’t be using it in production.
From Perimeter to Identity-Based Security
Old Way
The shared facility that Mia used had security guards at the gate.
Once inside, wearing a badge from Mia’s factory, you were trusted.
You could walk anywhere in the factory, touch any machine, and open any freezer.
What Needs to Change
The gate is no longer the problem. The problem is inside.
Mia’s robots have credentials.
They can access ingredients, adjust churner settings, move stock and read customer data.
If a robot’s credentials are stolen, or if any of them simply starts behaving strangely, it can cause damage without ever touching the gate.
And here’s the part that keeps me up at night: Mia’s robots talk to each other.
If an attacker compromises one robot, they can use it to influence the others.
The damage spreads through the swarm.
This is what makes AI security different.
It’s not about who trespasses. It’s about whether every robot is doing exactly what it’s supposed to do, every single time.
What This Means for You
You need to trust based on identity (which specific AI agent is this?), behaviour (is this what it normally does?), and context (does it have a legitimate reason to do this right now?)
A robot that suddenly starts accessing recipes it’s never touched before? That’s a red flag.
A robot that tries to move milk outside the facility at 3AM? That’s a red flag.
A robot telling another robot to quietly adjust churner settings for weeks without any valid reason?
That’s a bigger problem.
Location doesn’t tell you any of this. Identity, behaviour and context do.
A Quick Reality Check
I need to be honest with you.
This story is a simplification.
Mia’s ice cream factory isn’t a perfect analogy on how cloud infrastructure changes as we moved from on-premise to public cloud, and now utilising AI on public cloud.
It’s just a way to make a complex topic feel a bit more digestible.
In reality, cloud infrastructure is virtual, global and elastic.
AI isn’t just a swarm of robots walking around a factory churning ice cream. It’s often a complex system of models and agents working together to achieve goals.
The new billing unit based on tokens doesn’t replace the cloud infrastructure costs associated with compute, storage and network. It adds a layer on top.
But the core idea holds.
The old way of thinking about infrastructure no longer works.
The neat, separate layers we used to manage independently have collapsed, weaving into something more integrated, dynamic, and powerful.
And that’s what this new reality is all about: With AI - the new superpower - comes a fundamental shift in how we think about building & managing things.
What to Do on Monday
So if you’ve made it this far, thank you for putting up with my long-winded words.
And you’re probably thinking: “Okay, this is a nice story. But so what? What can I do?”
Here’s my take: Don’t panic. Don’t start overhauling your entire infrastructure.
That’s a one-way ticket to confusion and burnout.
Instead, start really small. Just pick an AI workload (or one of your planned AI workloads), look at it, and try to answer the following 3 questions:
Cost - Do you have real-time visibility into what you’re spending?
Traceability - Can you explain why your AI made a decision?
Security - Are you trusting based on identity, behaviour and context, not just location?
You’ll be surprised where the gaps are.
And here is one key lesson I’ve learned over the years: You don’t need to have all the answers. You just need to ask the right questions.
Mia didn’t redesign her whole factory overnight like Harry Potter waving a magic wand.
She started with one robot, saw what worked, what didn’t, and iterated from there.
One thing at a time.
You can do the same.
Your Turn
What has been your experience with AI? Are you seeing AI reshape your cloud architecture yet?
If you found this useful, subscribe to SkyeWorks. I write about technology the way it should be—simple, grounded, and human. No jargon, no ego, just things I’ve learned that may help you too.
See you in the next post.
