What is true about AI
Two engineers and AI vs an agency: the new economics
For decades the logic was simple: bigger job, bigger team. An agency won because it had more hands. AI broke that logic. When the tooling does the heavy lifting, output stops scaling with headcount, and the maths that justified the large, layered shop stops adding up. Here is the honest version of the new economics, in rand where it matters, with the source next to every number.
The fact that changed everything
Output used to scale with people. It no longer does. AI now resolves over 70% of verified real-world software bugs unaided (SWE-bench Verified), up from about a third in mid-2024. The grind that used to soak up a team’s week is increasingly automated. The leverage shows up at the extreme too: WhatsApp was sold for 19 billion US dollars in 2014 with just 55 employees. That kind of output-per-head is no longer a Silicon Valley outlier; the tools that made it possible are now on every desk.
Where an agency’s money actually goes
This is the uncomfortable part. A large share of a typical agency invoice never touches the work. Roughly 30 to 40% of it is account management and markup: the project managers, the layers of sign-off, the office, and the margin on subcontracted hands. That overhead is fair value when you need a big project formally managed across many people. It is poor value when you are paying it on top of work that AI now does in a fraction of the time.
A lean, engineer-led team carries almost none of that. There is no markup chain because the person you talk to is the person building it, and no account-management tax because there is barely an account to manage. More of every rand turns into output.
| Cost | Layered agency | Two engineers + AI |
|---|---|---|
| Who you talk to | Account manager, then a team | The people building it |
| Management + markup | About 30 to 40% of the invoice | Near zero |
| Speed on the build | Paced by team coordination | Paced by AI, fixed price |
| The heavy lifting | Billed human hours | AI does it (70%-plus on SWE-bench Verified) |
| Who owns the result | Often the agency, on retainer | You, handed over |
Illustrative, not a quote. The point is structural: a layered model spends a large share of your rand on coordination; a lean one spends it on the work.
”Small” no longer means “lower quality”
The instinct that big equals safe is out of date. In 2026 the quality ceiling is set by two things: the AI tools, and the skill of the people directing them. A small technical team has full access to both. And because there are no hand-offs, the people who scoped the work are the ones who build it, so intent does not leak between a strategist, a project manager, and three subcontractors. Fewer hands on the same outcome usually raises quality, not lowers it.
Why this is a competitive issue, not a budget one
This is not only about saving money. BCG (2025) found AI leaders grow revenue about 1.7x faster than laggards, and the leanest movers tend to be small and technical, because they can change direction in a day instead of a quarter. The pressure is structural: the average lifespan of a top-500 company is forecast to fall to about 12 years by 2027 (Innosight forecast). Speed of adaptation is becoming the whole game, and small beats big at speed.
The macro picture in South Africa points the same way. McKinsey (2025) put the African generative-AI prize at 61 to 103 billion US dollars a year, while 84% of large local corporates say they cannot find the skills they need (Xpatweb, 2025). The scarce resource is not headcount, it is people who can direct AI well. Whoever has that wins, regardless of size.
When the big team is still the right call
Honesty cuts both ways. A full agency is the right choice when the work genuinely needs managed scale: many parallel workstreams, large sign-off chains, regulated processes, or a brand that needs a big formal team standing behind it. If that is your situation, pay for the coordination, it is doing real work. For most small and mid-size builds, where speed, ownership, and direct access matter more than headcount, a lean technical team is the better buy.
That is the entire thesis behind how we work. Two engineers who actually ship, pointing the best AI on earth at your problem on a fixed price, with no markup chain and no hand-offs: that is Zaiq.
Where to go next
- AI in 2026: what is actually true, for South African business - the hub this page sits under.
- What AI just made obsolete (and what it did not) - including headcount as the measure of output.
- The AI benchmarks, explained for a normal person - why “AI does the heavy lifting” is a measured fact, not a slogan.
- What digital work costs in South Africa - the rand reference for every kind of build.
Questions people ask
Why can two engineers with AI out-ship an agency?
Because AI removes the labour bottleneck and an agency's price is mostly overhead, not work. AI resolves over 70% of real software bugs unaided (SWE-bench Verified), so a small technical team moves at a speed that used to need many people. With no account-management layer or markup chain, far more of what you pay turns into actual output.
How much of an agency invoice is overhead?
A large share. Roughly 30 to 40% of a typical agency invoice is account management and markup rather than the build itself, which is reasonable when you need a big project formally managed and poor value when you are paying it on top of work AI now does in a fraction of the time.
Is AI making agencies obsolete?
Not all of them, but it has removed their old advantage. Agencies used to win on capacity, having more hands. When AI gives a small team that capacity, the remaining question is whether you are paying for managed scale you need or for layers you do not. BCG (2025) found AI leaders grow about 1.7x faster, and the leanest movers tend to be small.
Does a small team mean lower quality?
Not in 2026. The quality ceiling is set by the AI tools and the skill of the people directing them, both of which a small technical team has full access to. With fewer hand-offs, the people who scoped the work are the ones who build it, which usually raises quality rather than lowering it.
What is the real cost difference in rand?
For comparable digital work in South Africa, a layered agency typically carries the project plus 30 to 40% in management and markup, while a lean engineer-led team prices closer to the work itself. The build-fee gap is real, and it widens once you add the two-year total of retainers and lock-in.
When should I still use a full agency?
When the work genuinely needs managed scale: many parallel workstreams, large sign-off chains, regulated processes, or a brand that needs a big formal team behind it. For most small and mid-size builds where speed and ownership matter more than headcount, a lean technical team wins.