What is true about AI
What AI just made obsolete (and what it did not)
“AI is going to replace everything” is a bad forecast, and so is “AI changes nothing”. The truth is a clean line. AI retired a specific layer of work, the repetitive, language-heavy, checkable layer, and left the rest standing. This is the honest map of what is genuinely gone by 2026 and what is not, with the source next to every claim, so you can place your own work on the right side of the line.
What AI actually made obsolete
The slow first draft. Producing a competent first version of almost anything written or coded used to be the bottleneck. It is now nearly free. AI resolves over 70% of verified real-world software bugs (SWE-bench Verified), up from about a third in mid-2024, which tells you the model is not just drafting prose, it is doing checkable engineering. The blank page stopped being the expensive part.
The week-long bug hunt. The manual slog of tracing a defect through a codebase is largely gone for common cases, for the same reason. The work moved from finding the bug to deciding which bugs matter and verifying the fix.
The blue-link habit. The old model of being found, ranking somewhere on a page of ten links, is fading fast. About 58.5% of US Google searches now end without a single click (SparkToro, 2024), and ChatGPT has roughly 800 million weekly users (October 2025). People increasingly get an answer, not a list. Being the tenth link on a page nobody scrolls is obsolete; being the source an AI quotes is the new game.
The static FAQ and the after-hours wait. A page of frozen answers and a “we will reply Monday” queue are hard to defend when a model can answer the same repetitive questions instantly, in plain language, at any hour.
| The old way | Status | The evidence |
|---|---|---|
| The slow first draft | Obsolete | Over 70% of real bugs auto-resolved (SWE-bench Verified) |
| The week-long manual bug hunt | Obsolete for common cases | Same: about a third in mid-2024, now 70%-plus |
| Ranking blue links nobody clicks | Fading fast | 58.5% of US searches end with no click (SparkToro, 2024) |
| The static FAQ / after-hours wait | Obsolete | ChatGPT about 800m weekly users (Oct 2025) |
| Headcount as the measure of output | Obsolete | WhatsApp sold for 19bn USD with 55 staff (2014) |
Sources cited in-text. Each row is a way of working the evidence shows has been overtaken, not a job that vanished overnight.
What AI did not make obsolete
Here is the part the hype skips. AI changed the cost of producing a draft. It did not change who has to be right.
Judgement and accountability. Any task where a confidently wrong answer is expensive and hard to catch still needs a human who owns it. A model graded at gold-medal standard at the 2025 Maths Olympiad (DeepMind) still cannot be the one accountable for your financial statements, because a benchmark rewards a clean win and your business punishes a confident mistake.
Taste, trust, and relationships. Deciding what is worth building, what is on-brand, who to trust, and why a customer should choose you: none of that is a checkable task, so none of it has been retired.
Original truth. AI can rephrase what exists. It cannot run your test, survey your customers, or be the named expert who stands behind a claim. As drafting goes to zero, this is precisely what gains value.
The test: will AI retire this task?
You do not need a forecast, you need three questions about any specific task:
- Is it repetitive?
- Is it mostly language or data?
- Can a human quickly check the output?
Three yeses and AI has probably overtaken it; hand it the grind and keep the judgement. If the task needs accountable judgement, or a wrong answer is costly and hard to spot, it is on the safe side of the line and a human stays in the loop. This is the same filter we covered when mapping what AI is genuinely good at.
What this means for an SA business
The opportunity is not “fire half your team”. It is “stop spending your scarce people on the layer AI retired”. McKinsey (2025) put the African generative-AI prize at 61 to 103 billion US dollars a year, while 84% of large South African corporates say they cannot find the skills they need (Xpatweb, 2025). The winning move is to let AI take the repetitive layer and point your actual humans at judgement, trust, and original work, the parts that survived.
The studios and teams that win the next few years are the ones who already moved their people up the stack. Two engineers who let AI do the grind and spend their judgement on the outcome: that is Zaiq.
Where to go next
- AI in 2026: what is actually true, for South African business - the hub this page sits under.
- The AI benchmarks, explained for a normal person - why the numbers behind “obsolete” are real.
- Two engineers and AI vs an agency: the new economics - what happens to the cost of work once the grind is automated.
- How to get found on AI search in South Africa - because blue links are no longer the game.
Questions people ask
What has AI genuinely made obsolete by 2026?
The slow first draft, the week-long manual bug hunt, the static FAQ page, and the habit of ranking blue links nobody clicks. AI resolves over 70% of real software bugs (SWE-bench Verified) and 58.5% of US searches now end with no click (SparkToro, 2024). The repetitive, language-heavy, checkable layer of work has been overtaken.
What has AI not made obsolete?
Judgement, accountability, taste, trust, and relationships. Any task where a confidently wrong answer is expensive and hard to catch still needs a human who is responsible for it. AI changed the cost of producing a draft; it did not change who has to stand behind the final call.
Will AI replace SEO?
It replaced the old version of it. With 58.5% of US searches ending without a click (SparkToro, 2024) and ChatGPT at about 800 million weekly users (October 2025), being found increasingly means being quoted by an AI answer, not ranking tenth on a page nobody scrolls. The goal moved from links to citations.
Does AI replace developers?
It replaces the slow parts of the work, not the developer. AI fixing over 70% of curated bugs (SWE-bench Verified) removes the grind, but someone still has to decide what to build, check the output, and own the result when it ships. The job shifted from typing code to directing and verifying it.
Is content writing obsolete because of AI?
Generic, undifferentiated content is. A first draft is now nearly free, so the value moved to what AI cannot fake: original facts, real testing, a named expert who is accountable, and a point of view. The floor rose, which means thin content lost its last advantage, cheapness.
How do I tell if AI will retire a task?
Ask three questions. Is it repetitive? Is it mostly language or data? Can a human quickly check the output? Three yeses and AI has probably overtaken it. If the task needs accountable judgement or a wrong answer is costly and hard to spot, it is on the safe side of the line.