Just because you can build it, does not mean you should.

AI has lowered the barrier to building software. It has not lowered the need for judgement, and that gap is where most of the value is won or lost.

I want to be clear from the start that this is not an argument against AI. My company uses it every day. It is an argument against building things for the sake of building them. That is a much older mistake, and all AI has really done is make it cheaper and quicker to make. The temptation has always been there. What has changed is that the cost of giving in to it has dropped to almost nothing, and when something becomes that easy to do, people stop asking whether they should.

There are thousands of apps and systems out there trying to solve a problem or add value, with hundreds of thousands more on the way. Very few of them properly understand the problem they claim to solve. That is the part that does not show up in a demo. You can generate a working interface, wire up a database, and ship something that looks finished, and still have built the wrong thing entirely, because the most difficult part was never the building. The difficulty comes from knowing what to build and why, and being honest about whether the thing in front of you actually moves a business closer to where it is trying to go.

Take automation as an example. It sounds good in every meeting, because who would argue against doing the same work with less effort. But automation that is not pinned to a clearly defined outcome does not remove work, it adds a new layer of operational overhead to manage, monitor and eventually untangle. Deploying a chatbot to "handle customer engagements" means very little if no one has first defined what those engagements actually are, what a good outcome looks like, and at which point a human needs to step back in. Without that groundwork, you have not reduced the work, you have simply moved it somewhere less visible.

An example I encountered recently (even though it dates back to 2023) was when I read about IKEA's chatbot, Billie, and how they approached the value from what AI could, and could not, offer them. Billie handles a large share of their routine, repetitive queries: order status, delivery windows, store hours, and the like. The interesting part of the story, to me, is not the chatbot itself but what IKEA chose to do with the queries Billie could not handle. When they looked at those, they found people asking for help with interior design and home planning, the kind of consultative conversation that no bot can meaningfully have. Rather than treating the freed-up call centre capacity as a cost to be cut, they retrained roughly 8,500 of those employees into remote design advisors. That new, human-led service went on to generate in the region of €1.3 billion (around $1.4 billion) in a single financial year. The automation did not create that value on its own. The judgement about what the technology was genuinely good for, and where people remained irreplaceable, is what created it.

The same trap catches internal tools. Dashboards look impressive in a demo, and they are satisfying to build, but many of them never end up supporting a single real decision. They get built because the data happened to be available, not because someone identified a decision that was being made badly for lack of the right information, in front of the right person, at the right time. A tool that does not change a decision is decoration, however well it has been engineered.

Then there is speed, which is probably the most seductive promise of all right now. Building something quickly is only valuable if it enables you to reach your goal at a faster pace than before. A footballer can score a goal within 5 seconds, but if that goal was scored past his own goalkeeper, the manager would understandably be extremely dissatisfied, regardless of how quickly the goal was scored. A fast proof of concept is genuinely useful, and we build them ourselves precisely because they let us test an idea before committing real resources to it. But if the thing you built does not move the needle, the time you spent building it quickly is still time wasted. It was simply wasted efficiently. Speed is a multiplier, and a multiplier only helps you if the thing it is multiplying was worth doing in the first place.

There is another part of this that is easy to underestimate if you have not spent years inside software engineering. A piece of software that looks like it is working is not the same as software that is actually sound. Someone building with AI still has to know what to look for beneath the surface: security, data protection, maintainability, scalability, the underlying architecture, the dependencies you are quietly taking on (most of which would be complete gibberish to a less experienced individual), and how the whole thing integrates with the systems and processes that already exist around it. These are not academic concerns. They are the difference between something that works on the day it is demonstrated and something that keeps working, safely, a year later when it is carrying real load and real data. AI can generate the code. It cannot carry the responsibility for that code once it is in production. That responsibility stays with the builder, with the service provider, every single time. This is precisely why I would argue that AI is at its most valuable when it is embedded inside a business process that is already well understood, and at its least valuable when it is bolted on so that someone can say they are doing something with AI.

There is also a local dimension to this that matters a great deal to us, because we build from South Africa, for a market with its own realities. Connectivity is the obvious one. Offline functionality is not a nice-to-have here; depending on what the software has to do, and where it has to do it, it can be the entire requirement. If a process runs in a rural area with unreliable connectivity, wiring in an AI agent that needs a constant connection to be useful might be completely redundant, or worse, an additional point of failure introduced into something that needed to be robust above all else. The right question is not whether we can add AI to something, it is whether the context even allows for it, and whether it would actually help if it did.

Small and medium enterprises are the other local reality worth naming. Many South African SMEs still run on manual workflows, and it is easy to look at that from the outside and conclude they are simply behind. More often there are real reasons for it: cost, infrastructure, the size and the skills of the team, the margins they operate on. Understanding the parameters around where a piece of software will actually live, and who will actually use it, can matter far more than knowing how to write or generate the code for it. Businesses here need solutions that fit their context, their budgets, their teams and their infrastructure, not solutions designed for a boardroom in a different economy. The opportunity in this market is not to chase whatever global AI trend happens to be loudest in a given quarter. It is to apply AI, and technology more broadly, practically and deliberately, to real operational problems that are genuinely worth solving.

All of which brings me to the kind of work we want to do, and the kind of businesses we want to do it with. We want to partner with companies and people who are curious, open to innovation, and genuinely willing to improve the way they operate, while remaining honest about the complexity that comes with it. That combination matters more than it might sound. The willingness to change and the respect for how difficult good change actually is have to travel together; either one on its own tends to end badly.

For us, AI is a tool. It is not a one-stop shop that quietly solves every problem, it is something that makes us more productive while we keep full responsibility for the work we deliver. We use it as part of a broader engineering and problem-solving toolkit, not as a shortcut that lets us skip past the expertise. There is a real difference between those two things, and our clients feel that difference in what eventually gets shipped, and in how it holds up long afterwards.

The future does not belong to the businesses that adopt AI blindly, and it does not belong to the ones that ignore it either. It belongs to the businesses that learn to use it with purpose. That, in the simplest terms, is what Claer & Volker is here to do. We partner with businesses that want to use technology intelligently. AI is part of that toolkit, but our focus has never changed: understand the business problem, design the correct solution, and build systems that create real value.