Opinion, from engineers who have seen what happens when you deploy on a Friday.
The robots were supposed to take our jobs by now. We checked. The jobs are still here.
Some even came back.
The Apocalypse Missed Its Deadline
In September, The Economist ran The jobs apocalypse is postponed. An AI jobs boom is here. The short version: "no sign of it yet".
- America added 162,000 jobs in August. Unemployment sat at 4.1%.
- AI created roughly 1 million US jobs, by The Economist's estimate. Layoffs blamed on AI since mid-2023: about 200,000.
- Engineers, developers, mathematicians and data scientists added some 730,000 jobs above trend since 2022.
- Young workers are fine. The gap between unemployment at 20–24 and the overall rate is near a multi-decade low.
It isn't painless. Customer-service jobs are down about 10% since January 2023, administrative assistants about 15%. Real people, real pain. But the net number points up, not off a cliff.
The Boomerang
Here's the plot twist. Plenty of companies didn't wait for the data. They cut first and planned to let the agents sort it out.
The agents did not sort it out.
- Klarna replaced a big part of customer service with AI, then said quality had dropped and started hiring humans again.
- Commonwealth Bank of Australia cut 45 roles for a voice-bot. The bot created more work for humans. The bank reversed the cuts and called it an "error".
- 55% of leaders who made people redundant because of AI admit they got those decisions wrong (Orgvue, 2025).
- Gartner expects half of organisations planning big AI cuts in customer service to drop those plans by 2027 (Gartner, 2025).
Picture This: An Accounting Firm Discovers Production
A made-up firm, but you'll recognise it.
Q1. The board reads a slide that says "agentic". The bookkeeping team goes. Agents arrive: one for invoices, one for reconciliations, one for month-end close.
Q2. The agents talk to each other. Nobody reads the conversation, like a Kafka topic with no consumer: everything recorded, nothing understood. A rounding error in the invoice agent becomes an input to the reconciliation agent. (C++ developers call this undefined behaviour. Accountants call it a restatement.)
Q3. Month-end close takes longer than before. The auditors have questions. The agents have confident answers. Different ones.
Q4. The firm hires bookkeepers. Often the same ones. This time the job ad says "experience with AI tools".
Anyone who has run a trading system knows this movie. You never switch on an algorithm without a kill switch, limits and someone who understands the market watching it. Speed without supervision isn't automation. It's a faster way to lose money.
Engineers have a name for the firm's setup: an open loop. The system acts, nobody measures the result, nobody corrects it. The bookkeepers were the feedback loop. Remove them and errors don't stop, they just stop being noticed.
Here's the insight we keep coming back to: when doing gets cheap, checking gets valuable. Agents made the doing cheap. The people who know what "correct" looks like just got more expensive to lose.
The holy grail was never "no people". It's people who know the work, running agents that do the boring parts.
What the Job Ads Actually Ask For
We read job ads for a living (our machines do the reading, we drink the coffee). Over the six months to mid-September, we looked at 58,021 AI-related postings across 45 roles.
- Finance managers: 98% of AI-related postings ask for AI tools literacy, 86% for Excel automation. Only 0.2% are "core AI" jobs. Translation: finance people, with AI. Not AI people.
- Financial analysts: 92% ask for AI tools literacy, 42% for prompt engineering.
- Executive assistants: 82% ask for AI tools literacy. Calendar automation shows up in 35%.
- Customer success managers: 62% ask for prompt engineering. That's a fancy name for explaining clearly what you want. You've done that before.
That's the data. Here's our read: employers want the person who knows the work, plus the tools. The tools are the easy half.
Your Move This Week
No bootcamp required. Twenty minutes and a coffee will do.
- Write down your edge cases. Three things your job taught you that no manual mentions. That's what agents get wrong, and what employers pay for.
- Check your role. See what employers ask for in finance, analysis or customer success, or browse all AI-enabled jobs.
- Add one tool to real work. Pick one skill from what employers want and use it on Monday's task. The first try will be clumsy. Ship it anyway (not on Friday).
The apocalypse is postponed. The rehiring has started. Be the person they call back, or better, the one they never let go.
Figures: The Economist, The jobs apocalypse is postponed. An AI jobs boom is here, 4 September 2026; Orgvue, Gartner, Klarna and Commonwealth Bank as linked. Role figures: SlashHash analysis of 58,021 AI-related job postings across 45 roles, 180 days to 14 September 2026, from our AI career guides.