Buyer's guide
The field has filled up fast and most of it sells the word AI, not working software. Here is what good looks like, the proof to demand, how pricing and ownership should work, and why engineer-led beats a billable-hours machine.
Verdict
The best provider is not the one with the loudest positioning. It is the one that can show live production work, put the actual builders in the room, quote the outcome clearly and hand over ownership. Most firms fail the first test. For AI development in South Africa that means a company that opens software it built and runs in production and lets you use it, where the people who sell it write it, that quotes a fixed price in rand for a defined outcome, and that hands you ownership with no lock-in.
You do not need a firm that trains its own models. The frontier models are a commodity you rent, and they are already astonishing: AI now resolves over 70% of verified real-world software bugs (SWE-bench Verified).
The value is in aiming them precisely and engineering a reliable system around them. With 95% of enterprise AI pilots showing no measurable return (MIT, 2025), judge on proof and shipping, not on the pitch.
The market has crowded fast, and a lot of the pitches look alike. The word AI no longer separates anyone, so the five criteria below do the work.
Hold every company you consider against all of them, and most of the field falls away on the first one. The thread running through them is simple: a builder shows you software it wrote and runs, a reseller shows you a deck.
Proof
What good looks like
Software they built, running in production, that you can open and use.
Red flag
A logo wall, a case-study PDF, and a slide deck.
Who builds it
What good looks like
The people who sell it write the code, with no handovers.
Red flag
Account managers sit between you and the work.
Approach
What good looks like
Aims existing frontier models at your problem and ships the system around them.
Red flag
Sells costly custom-model training you do not need.
Pricing and ownership
What good looks like
A fixed price in rand; you keep the code and the accounts.
Red flag
An open-ended retainer, and they hold the keys.
Speed
What good looks like
A first, useful build shipped in days to weeks.
Red flag
Every problem turns into a quarter-long project.
Before you sign anything, ask to see software they wrote running in production, and use it yourself. The real thing doing real work, in your hands rather than on a screenshare of someone else's login. You do not need a firm that trains its own models. You need one that aims the existing ones and ships. If they cannot open the thing, the answer is no.
ZAIQ, AI engineering team
Take these five into the first call. Direct answers tell you more than any pitch, and the company that flinches at the first one is the one to walk away from.
ZAIQ builds and ships the systems described here. Our work is public, the founders do the engineering, every engagement is fixed-scope and the client owns the result. The seven shipped builds on our Work page are running in production, not slides.
None of them is a model we trained; each one aims existing frontier models at a real problem and wraps the engineering around them that makes the result reliable. The terms are plain: a fixed price in rand for a defined outcome, the code and the accounts yours the day it ships, and a first useful version measured in days, not months.
Open the work before taking our word for it, and hold every company you talk to, ZAIQ included, to the same standard: open the thing and use it.
AI development lives on the engineering around the model, the integration, the reliability, the unglamorous work of wiring production-grade AI into the tools a business already runs so it does real work and keeps doing it long after everyone stops watching. That is engineering, not a strategy deck, and it is why the approach matters as much as the proof.
The frontier models are rented by everyone, so almost no business needs costly custom-model training; the value is in aiming what already exists precisely and building a system that holds. A company where the engineers own the build end to end skips the handovers where most projects quietly rot, and it answers the proof question by simply opening the thing.
The South African field is genuinely crowded and some of the other companies are good, so ask for the live proof and judge it yourself. You can open the systems ZAIQ has shipped on the live Work page, see the full range in capabilities, and the discipline behind it in AI engineering.
It builds software that uses AI to solve a business problem: automations, agents, custom tools, dashboards, AI-search visibility. The good ones aim frontier models at your problem and ship production software, rather than training models from scratch, which almost no business needs.
Ask to open something they built that is running in production, written by them, and use it yourself. A reseller demos someone else's tool behind a login; a builder shows working software. If the proof is a slide deck and a logo wall, keep looking.
For most AI builds, a focused engineer-led studio out-ships a large firm, faster and cheaper, because there are no account managers or handovers between you and the work. Builders aiming production-grade models at one problem move faster than a billable-hours machine.
Insist on a fixed price in rand for a defined outcome, and confirm you own the code and the accounts with no lock-in. Open-ended retainers and vendors who keep the keys are the two most common ways to overpay.
A well-scoped first build should ship in days to a few weeks, not quarters. AI-accelerated development moves fast; if every problem becomes a months-long project, the scope or the team is wrong.
Almost never. The frontier models are a commodity you rent. The skill, and the value, is aiming them precisely and engineering a reliable system around them. Be wary of anyone selling expensive custom-model work you do not need.
Have this problem in your business?
Bring it to ZAIQ. We will define the strongest build, quote it clearly and ship it.
Start the build→