David Noakes, Software Architect at Retro Rabbit/SmarTek21.
AI is transforming every aspect of the software development life cycle, but because the technology delivers such professional-looking outputs, organisations risk falling into an ‘AI trap’ and overlooking key factors in development.
This is according to David Noakes, Software Architect at Retro Rabbit/SmarTek21, a leading African design and engineering software consultancy. Noakes says: “AI can improve and speed up all aspects of the software life cycle – from writing specs to building the code to testing it. But organisations should not just blindly trust it and fall into the traps of neglecting the human connection and overlooking major domain issues.
“An LLM-assisted document can now be produced from little more than a strong idea, a decent prompt and some context,” he says. “AI is fluent in ‘plausible’. It's very good at producing answers that look convincing, but it can't tell you which of those answers is the correct one for your business, your customers and your regulator. That judgment comes from people who understand the domain deeply.”
Noakes notes that the scarce, valuable skills needed today are not the ability to generate options, but rather the ability to know which option is the right one. He says: “For example, the specification document that's meant to capture what a project needs. A spec put together with AI can be genuinely hard to argue with: it’s comprehensive, well-structured and thoroughly considered. It may look polished, complete and confident – all while quietly missing the very areas that matter most. It could omit all the nuances that make or break the project. In the past, a document like this was earned. It came from real domain knowledge and lengthy sessions with key stakeholders, so even when it took time to pull together, it tended to capture the most important and most nuanced parts of the problem early. The traditional spec document may have looked rough around the edges, but it covered what counted."
He adds: “A signed off spec captures what was said in the meeting, but the real problem usually lives in what was meant. It’s the unwritten stuff, the human stuff: instincts, exceptions, that look of concern from a key stakeholder. That quick conversation with the quiet, sharp developer who didn’t speak up, but filled half a notebook.
“It takes extensive experience and human skill to read these situations and to spot when a spec, or a whole project, has drifted from the real problem,” Noakes says. “This matters more now because of AI, not despite it. The building phase used to be slow, and that slow pace was a safety net: sprint after sprint of development gave you time to notice something was off and correct your course. AI removes that safety net, it builds the spec fast, and it looks finished far sooner than before. For many industries, that might not sound like a problem. But in highly regulated ones like insurance or banking, ‘looks right but isn't’ is exactly the kind of thing that turns into an expensive problem down the line.
“AI can be innovative, and it can unlock people to be more creative, brainstorm against it and use it as a sounding board. But it’s important that it doesn't take over the process – that's the trap,” he warns.
Noakes says Retro Rabbit/Smartek21 has embraced AI but depends on human skill and experience to harness it effectively. “It’s the new paradigm in how software is built, and mastering it is essential. But mastery isn’t prompting, it’s the discipline around it. We place a lot of emphasis on creating standards and context within the project, the harness that ensures consistent quality of the results,” he says.
“When anyone can build it, knowing what to build is the whole job,” he emphasises. “When organisations work with a software consultancy, they’re not paying for faster code – they are paying for people who understand their world well enough to protect them from costly mistakes. People with decades of experience who can guide them through an industry changing faster than anything we've seen before,” Noakes concludes.
