AI engineering · Cape Town and the UK
Start with one piece that works. We redesign the work, build the software and AI agents behind it, and set it up inside your own systems with proper controls. Then we take the next piece.
You'll speak to one of the two founders.
The shift
Your team probably uses AI already. Someone drafts emails with ChatGPT. Someone else built a handy tool over a weekend. A partner pastes documents into whichever assistant happens to be open.
Each of those people is quicker. The business, taken as a whole, works much as it did last year. The tools don't talk to each other. Nobody else can maintain what one person built. And nobody is quite sure where the documents went.
Closing that gap is the work. It means deciding which jobs stay with people and which go to software or an AI agent, where the data lives, and who approves what. A business that has done this is what we mean by AI-native.
We think firms built this way will pull ahead over the next few years. 2032 is the horizon we steer by. That's a hypothesis, and we label it as one.
Our longer-range hypothesis is that by around 2032, businesses genuinely built around AI will hold a serious advantage, and firms that only added a few tools will struggle to keep up. We can't prove that, and we treat 2032 as a direction to steer by. AI changes every few months. Nobody knows exactly what an AI-native business will look like, and that includes us. So we don't sell a finished design. We help you keep moving towards one, a working piece at a time, and we check the direction against what is actually working.
We haven't picked a single industry. We suit firms in a particular situation:
Size matters less. We're set up for firms from a handful of people to a couple of hundred. [confirm: size band]
What we believe
We wrote our thinking down before we wrote this site, and we keep testing it. Some of it we hold firmly. Some of it is still a hypothesis. Each idea carries a label so you can tell which is which.
Select an idea to open it.
We think of a business that runs on AI as six layers working together. People own judgement, relationships, strategy, the exceptions and the accountability. AI agents take on research, analysis, sorting and repetitive knowledge work. Software handles what must behave the same way every time: integrations, transactions, actions in your systems. Data gives all of it context and memory. Governance sets who can do what, keeps a record, and puts a person in front of decisions that matter. Continuous improvement keeps finding the next thing to redesign. Plenty of AI projects touch one layer and stall at the next. We plan across all six from the first conversation.
Open an AI assistant and you see a box, a prompt and an answer. We see a longer chain: models, agents, APIs, workflows, orchestration, context, retrieval, software, cloud, infrastructure, automation, business processes, data, systems, governance, operations, and at the far end, economic leverage. Between us we have worked on most links in that chain, from cloud platforms and payment systems to running a software business. [confirm] So we start with how your business actually works, and treat AI as a new layer to redesign it with. Writing a good prompt is the easy part.
We can picture where this might go. Services turn into repeatable packages. Packages become shared platforms. Later, perhaps, AI agents do paid work across company lines, even for other businesses. That last idea is speculative, and we don't sell it. Our rule is to prove phase one before building anything for phase five. Phase one is plain: solve real problems for paying clients, measure what changed, and let the evidence choose the next step. A line from our own notes keeps us honest: being early isn't the same as being right.
Our longer-range hypothesis is that by around 2032, businesses genuinely built around AI will hold a serious advantage, and firms that only added a few tools will struggle to keep up. We can't prove that, and we treat 2032 as a direction to steer by. AI changes every few months. Nobody knows exactly what an AI-native business will look like, and that includes us. So we don't sell a finished design. We help you keep moving towards one, a working piece at a time, and we check the direction against what is actually working.
Charging by the hour tells you our value is our time, and we don't think it is. So we sell defined pieces of work at a fixed price, agreed in writing before we start, and a monthly partnership for what comes after. We'd like to be judged on what each rand spent with us gets you. Over time we want more of what we deliver to be reusable software and agents, which should stretch that further. One honest caveat: a fixed price still rests on our estimate of the effort. If the estimate is wrong, that's our problem.
On day one we map where the time goes in one part of your business, and agree two or three numbers with you: hours spent, say, or errors caught. We write them down before we change anything. In the first weeks, one workflow goes live and your people use it on real work. [confirm: timing] At the end of the first month, we measure against the day-one numbers and choose the next piece together. Our own plan once said "week one". Some work is that quick and some isn't, so we commit to the baseline and the working piece, and agree the dates with you.
AI makes some things practical that never were at human scale. It's tempting to call that revolutionary. We'd sooner write it down. Every piece of work starts with a before sheet: how many people touch the task, the hours, the manual steps, the systems involved, the error rate, the throughput. Afterwards we fill in the same sheet again, and add what's new: the steps an agent now does, where a person still approves, how long it takes now, how good the output is. You keep both sheets. If a number didn't move, you'll see that too.
We run our own firm the way we'd help you run yours. Our meetings are transcribed and filed, and the raw transcript outranks anyone's summary. Every belief we hold about our market sits in a register, with a confidence level and the evidence for and against it. An AI agent keeps our client records up to date while we work. Our internal tools sit behind an identity gate, with nothing open to the public internet. Nothing we write reaches a client until three separate AI reviewers have read it. We're two people, so this proves little about a firm of sixty. It does show you what we hold ourselves to.
What we do
Most firms start small and add pieces as they go. Every price is fixed and agreed in writing before we begin.
Half a day with your team, on your own work.
[confirm: price]
Book a working sessionOne workflow live, and the numbers to judge it by.
[confirm: price]
Ask about a first workflowA private place for your AI work, inside your own cloud account.
[confirm: price] plus your cloud costs
Ask about your own environmentWorkflows, agents and integrations that make AI part of the day’s work.
From [confirm: price] per piece
Talk about a buildAI capability on call, without hiring for it yet.
From [confirm: price] a month
Ask about a partnershipNo free pilots. The first conversation costs nothing and should be useful on its own. After that, work starts small, paid and fixed-price, so we both find out quickly whether it’s worth doing.
Charging by the hour tells you our value is our time, and we don’t think it is. So we sell defined pieces of work at a fixed price, agreed in writing before we start, and a monthly partnership for what comes after. We’d like to be judged on what each rand spent with us gets you. Over time we want more of what we deliver to be reusable software and agents, which should stretch that further. One honest caveat: a fixed price still rests on our estimate of the effort. If the estimate is wrong, that’s our problem.
How we work
We map where the time goes in one part of your business and agree the numbers that matter. We write them down before we touch anything.
One workflow goes live. Your people use it on real work, and we adjust it with them.
The people who’ll use it help shape it. If someone has already built a tool of their own, we fold it in so others can rely on it.
We fill in the same numbers again and show you what moved, and what didn’t.
We choose the next piece together. Each one is priced, built and measured the same way.
On day one we map where the time goes in one part of your business, and agree two or three numbers with you: hours spent, say, or errors caught. We write them down before we change anything. In the first weeks, one workflow goes live and your people use it on real work. [confirm: timing] At the end of the first month, we measure against the day-one numbers and choose the next piece together. Our own plan once said “week one”. Some work is that quick and some isn’t, so we commit to the baseline and the working piece, and agree the dates with you.
AI makes some things practical that never were at human scale. It’s tempting to call that revolutionary. We’d sooner write it down. Every piece of work starts with a before sheet: how many people touch the task, the hours, the manual steps, the systems involved, the error rate, the throughput. Afterwards we fill in the same sheet again, and add what’s new: the steps an agent now does, where a person still approves, how long it takes now, how good the output is. You keep both sheets. If a number didn’t move, you’ll see that too.
Security and data
The first question a careful firm asks is what happens to confidential documents once AI is involved. Here is our answer.
When it matters, we build inside your own cloud account, in a South African region [confirm], and your data stays there. If we host something for you, we tell you exactly where it runs and who can reach it. [confirm: region of CloudSprint’s own hosting]
People sign in with the accounts they already use, and access follows their role. Nothing we set up is left open to the public internet. We run our own tools the same way.
Illustrative roles. Only the threads a role may reach are lifted, and the shuttle passes only through that opening.
Each agent’s actions are logged: what it read, what it did, and who approved it.
Illustrative sample. Not real data.
Anything that sends, pays, deletes or commits you to something waits for a person to approve it, unless you choose otherwise. [confirm: default]
The row is waiting for a person to approve it.
We work to POPIA, and we sign a data processing agreement before we handle any personal data for you. [confirm]
Lorenzo has spent more than ten years on cloud platforms where this mattered, including payment infrastructure built to PCI DSS Level 1. [confirm: employer naming and wording]
How we run CloudSprint
We’re two people, so none of this proves what will work in a firm of sixty. It shows what we hold ourselves to. Ask on a call and we’ll share our screen.
We run our own firm the way we’d help you run yours. Our meetings are transcribed and filed, and the raw transcript outranks anyone’s summary. Every belief we hold about our market sits in a register, with a confidence level and the evidence for and against it. An AI agent keeps our client records up to date while we work. Our internal tools sit behind an identity gate, with nothing open to the public internet. Nothing we write reaches a client until three separate AI reviewers have read it. We’re two people, so this proves little about a firm of sixty. It does show you what we hold ourselves to.
Ask for a walkthrough [confirm: offer to screen-share]
Who we are
We started CloudSprint in early 2025. We’d worked together before, and we still play tennis together. [confirm: tennis line] Demian leads on clients, strategy and the people side. Lorenzo leads on platforms, data and security. Both of us work on every engagement. [confirm]
Co-founder and CEO, UK [confirm: city]
Finds the problems worth solving and makes sure what we build suits the people who’ll use it. A self-taught developer who built, and still runs, his own software business. [confirm: business name and wording]
Co-founder and CTO, Cape Town
More than ten years building and running cloud platforms, including payment infrastructure to PCI DSS Level 1 and AI document pipelines. Microsoft Certified DevOps Engineer Expert. [confirm: wording; certification current]
Lorenzo Jenecker, Cape Town.
hello@cloudsprint.co.za
Demian Hauptle, [confirm: city].
hello@cloudsprint.co.uk [confirm: mailbox]
Our business development is in the UK, one hour behind Cape Town in summer and two in winter.
Start a conversation
A first call takes about thirty minutes. [confirm] You tell us where the time goes and what worries you. We tell you plainly whether we can help, and what a first step would cost. If we can’t help, we’ll say so, and where we can, we’ll point you to someone who can.
Or email hello@cloudsprint.co.za
We usually reply within one working day. [confirm]
About this build
This is a first slice of Concept A, the Loom and the Woven Country (11 §2 in the spec). It exists to judge one thing: whether a procedural, real-time scene can look and feel like a finished piece. The contact form is not built here: the call to action goes to the flat contact page.
For the previous attempt, a lighthouse lens, see aesthetic-full-concept/. It is left exactly as it was.