Building Got Cheap. Deciding What to Build Didn’t.
Everyone is selling AI. Very few are helping businesses decide what it should actually do. Why good advice matters more than ever, and a two-minute way to find out where your business stands.
The loudest message in business right now is that AI is easy. Just plug it in. No skills needed. Set it up in minutes, and let it run your admin while you sleep.
Some of that is true. Building got cheap. A tool will draft the email, summarise the report, sort the inbox and write the first version of the quote in seconds, for the price of a subscription. Things that used to take a developer, a designer and a month now take an afternoon.
What didn’t get cheap is deciding. Which of your tasks should AI touch? How far should it go? Who checks its work? What must never go into it? Those questions don’t come with the subscription, and most of the people selling AI would rather you didn’t ask them.
That gap, between how easy AI is to start and how hard it is to start well, is where most businesses are stuck. They tend to get stuck in one of two ways.
Rushing in
The first way is to believe the promises and go.
It usually starts with good intentions. Someone on the team discovers a free AI tool, it saves them an hour, and soon half the office is pasting work into it. Nobody decided that. Nobody wrote down what may and may not go in. And because it is free and it is fast, nobody asks where the information ends up.
This is not a small-business problem. In 2023 engineers at Samsung, one of the most sophisticated technology companies in the world, pasted internal source code into ChatGPT while trying to fix bugs. Samsung responded by restricting generative AI on company devices altogether. If it can happen there, it can happen in an office of twelve, with a client list instead of source code.
Then there is what AI says on your behalf. In 2024 a Canadian tribunal ordered Air Canada to compensate a customer after the airline’s website chatbot gave him a wrong answer about its bereavement fares. The airline argued that the chatbot was responsible for its own words. The tribunal disagreed: it makes no difference, it said, whether the information comes from a static page or a chatbot. If your AI tells a client something, your business said it.
Here in South Africa, the question people ask is whether there is an AI law yet. There isn’t: the draft national AI policy published for comment in April 2026 was withdrawn soon afterwards. That is not a free pass. POPIA already applies to every piece of client and staff information that goes into an AI tool, and it already says a person may not be subject to a decision about them based solely on automated processing when that decision significantly affects them. Hiring, pay, credit, cover: those stay with a person.
And there is a quieter risk that the easy promises skip entirely. When Harvard Business School and BCG gave 758 consultants access to AI, those working on tasks AI is good at finished 12.2% more tasks, 25.1% faster, at 40% higher quality. But on a task just outside what AI does well, the consultants using it got the right answer 60–70% of the time, against 84% for those working without it. AI made capable people worse at that task, and they didn’t notice. The line between the tasks where AI helps and the tasks where it hurts is real, and it is not where you’d guess.
None of this is a reason to avoid AI. It is a reason to decide before you deploy.
Standing still
The second way to get stuck is the opposite: not starting at all.
Every week brings a new tool, a new model and a new webinar promising this one changes everything. The budget is small. The business can’t afford to spend it on the wrong thing, and nobody inside has the time to work out which thing is the right one. So the sensible-feeling decision is to wait until it settles down.
It won’t settle down. And waiting has a cost of its own.
Competitors who have found the one task where AI genuinely helps them are already quoting faster, answering sooner and catching more of their own mistakes. And, more uncomfortably, standing still rarely means your business isn’t using AI. It usually means your people are using it anyway, on their own phones and free accounts, without rules. Not deciding doesn’t remove the risk. It just hides it.
If you are reading this thinking we’re falling behind, or I wouldn’t even know where to begin, you are in good company. Those are the two most common places we meet business owners, and both have the same way out.
Good advice is now the scarce part
When building gets cheap, the value moves to deciding what to build. That makes good advice more valuable than it has ever been, and harder to find among the noise.
Here is what we think it looks like, and how we work:
- It starts with your people and your tasks, never with a tool. The tool follows from the task. Research from BCG puts it bluntly: about 70% of what makes AI pay off is people and processes, and only 10% is the technology itself.
- It judges tasks, not jobs. “Admin” is not something AI can do. “Capture supplier invoices into the accounting system” might be.
- It says how far AI should go, task by task. Sometimes AI only assists a person. Sometimes it drafts and a named person approves. Sometimes it runs on its own and is measured. How far depends on what a mistake would cost, and decisions about people always stay with a person.
- It buys before it builds. Often the AI you need is already inside software you pay for. A custom build comes last, not first.
- It proves things small. Anything new starts with the smallest step that shows it works, before you commit to the whole thing.
- It lets you decide. Every recommendation sits on one page with the options, including doing nothing, and you sign it.
- It comes back. Every 90 days we measure what changed, and only step a task up when it has earned it.
Bad advice sells you a tool. Good advice tells you which of your tasks deserve one.
Where to start: two minutes
We built the Business AI Quiz for exactly the two places people get stuck.
It opens with one question: why are you here? One door is for ”I’m worried we’re falling behind.” The other is for ”I don’t know where to start.” Each asks a short set of plain yes, no or don’t-know questions about how your business actually works today. It takes about two minutes.
You see your result straight away, with no email needed:
- Your Business AI Readiness Score, from 0 to 100, with a plain-language name for where you stand.
- The gaps that matter most, each with a fix you can act on.
- A safety check: for example, whether client or staff information may be going into AI tools without anyone checking where it ends up. If it is, that comes first.
If you want more, you can have the full report emailed to you, with every gap we found, where AI could start in your business, and one safe thing you can do today.
It is a starting point, not a diagnosis. But two minutes is enough to turn “AI is something we should probably look at” into “here is where we actually stand”.
And after the quiz, a clear path
If the result raises questions, the next step is a 30-minute call. You tell us about your business, and you hear how we would build your plan. That’s all it is.
If it makes sense to go further, we run a fixed-price analysis, usually about three weeks. We look at how your work really gets done, with you and your team. Every task you name gets the same five questions: Does it pay? Is it allowed? What kind of work is it? How far should AI go? How new is it to you? Only the tasks that pay, and are allowed, make the plan. Your team’s worries are part of the work too, asked anonymously, because people rarely tell the owner they’re afraid of being replaced.
You end with your AI plan: where you are, every task worth doing in the order worth doing it, and one page per decision, signed by you. Then we make it work, cheapest route first: training your people on what you already have, setting up a tool, or building something only when nothing else fits. And every 90 days, we measure and go round again.
The plan is yours. You can run it with anyone. We think you’ll want to run it with us.
Who we are
Loopworks is a software studio in Cape Town. We build and run live AI products, and we run our own business on AI, with a team of AI agents working alongside us on research, analysis, design and development. We use what we recommend, every day, which is how we know where it helps and where it quietly doesn’t.
The promise, honestly
The promise you’ll hear everywhere is that AI is easy. Here is a more useful one: using AI is easy; using it well is a decision. Rushing in and standing still both feel safer than they are, and the way between them starts with knowing where you stand.
Take the Business AI Quiz: two minutes, and your result straight away → quiz.loopworks.io
Sources
- Samsung’s 2023 restriction of generative AI after engineers uploaded internal source code: reported by Bloomberg; see TechRadar and CSO Online.
- Moffatt v. Air Canada, British Columbia Civil Resolution Tribunal, February 2024; see AI Business and Manatt.
- South Africa’s Draft National AI Policy, published for comment in April 2026 and withdrawn: Cliffe Dekker Hofmeyr, GoLegal. POPIA section 71 on automated decision-making: Pinsent Masons.
- Dell’Acqua et al., Navigating the Jagged Technological Frontier, Harvard Business School and BCG, 2023; figures as summarised by co-author Ethan Mollick.
- BCG’s 10-20-70 rule: BCG.