What AI in your product really costs, when ordinary code does the job better, and where a thin layer of AI still earns its place.

It’s tempting to put AI into every feature customers see. Each one quietly adds cost and slows down how fast you can improve the product.

We asked Magic Safari founder and CEO Philip Joubert, whose team uses AI heavily behind the scenes, why he keeps it out of the product as much as possible…

Philip’s rule isn’t no AI. It’s just enough AI. Use it where nothing else will do, and use ordinary code everywhere else. His team uses AI constantly to build the product; the product itself contains as little as they can manage.

“Ask what problem you’re really solving, and whether AI is really the best solution.”

Ordinary code is easy to check: it’s either right or wrong. AI output isn’t, so it needs ongoing testing to catch bad answers; its prompts need upkeep, and the models it relies on get retired and replaced. The good models are also expensive to run. All of that slows down how quickly you can change and improve the product.

Magic Safari wanted customers to search for very specific trips, such as “lodges in Sabi Sand with heated private plunge pools in-room.” The obvious answer was an AI search that could take any request. The real problem, though, was matching people to lodges with the features they wanted, and that turned out to need much less AI.

Their AI search worked, but each search took around 20 seconds and some results were, in Philip’s words, bizarro. Rather than spend time making it faster, they built a large set of ordinary filters: location, room features, pools and so on. Filters are fast, predictable and easy to check.

They still use AI, but only for the one job filters can’t do: understanding what a customer typed. A thin AI layer turns the search phrase into the right filters, and the filters do the rest. AI handles the messy human input; ordinary code handles everything after it.

The result was an instant search that is almost free to run and far easier to maintain than the full AI version. Those three measures (how fast it is for the customer, what it costs per use and how much work it takes to keep running) are the fair test for any AI feature you’re considering.

Your product gets faster, cheaper and more reliable, and your team keeps the ability to change it quickly. The AI goes where it actually adds something, and your customers get answers in a second instead of twenty.

Philip’s full keynote from The Open Letter Cape Town 2026 is available to members inside the Founder Collab, where he goes further than we could cover here:

You’ll also get access to 40+ masterclasses from SA founders and operators on sales, fundraising, UX, paid media and more inside The Founder Collab.

Get more SA tech and business news, tips and business-building workflows and subscribe to The Open Letter.