The jobs AI-pocalypse is here: now what? An AI hockey stick moment, a token-based economy and Veo 3 as the scary near-future of film?
The jobs AI-pocalypse is here
Finally, someone from Big Tech has decided to come clean about how the race to embed AI into everything will decimate white-collar workers. Anthropic CEO Dario Amodei said this week that fears about AI taking over jobs are not only valid but already becoming reality. Amodei reckons AI could eliminate up to 50% of all entry-level white-collar jobs within five years, potentially pushing US unemployment as high as 20%.
“Most of them are unaware that this is about to happen,” Amodei said. “It sounds crazy, and people just don’t believe it.”
I warned back in early 2024 about the way AI will reshape office work, and the reality is quickly catching up with that prediction. If you break down what most office workers actually do – finding information, analysing data, writing reports or code, and implementing processes – AI already does these tasks faster, cheaper and increasingly better. Back then, my main argument was that it didn't matter that AI was still occasionally making things up. The real danger to office jobs was the widespread perception that AI could replace people, and that mind-virus perception has now infected both the public (e.g. Elon Musk’s AI-cost-cutting DOGE nonsense) and private sectors.
The data back up these concerns. Goldman Sachs estimates that generative AI could impact around 300 million jobs worldwide over the next decade, automating as much as 25% of white-collar tasks. The World Economic Forum expects AI to eliminate 83 million jobs globally by 2027 alone, particularly targeting clerical roles such as data entry and administrative support. This belief in AI's capabilities has already infected cost-conscious managers, who are eager to boost profits and please their boards with dramatic cost reductions.
Despite endless talk about AI ‘augmenting’ rather than replacing jobs, automation is already quietly slashing roles across multiple sectors. Jobs like translators, copywriters, customer service agents, data analysts, gig workers and junior coders are firmly in AI's cross hairs. Companies like IBM are openly freezing hiring for roles AI can handle, targeting about 7,800 back-office jobs. Meta's Mark Zuckerberg admitted that by the end of 2025, AI might do the work of mid-level coders, potentially wiping out thousands of junior developer positions. Even the creative fields aren't safe, as AI content-generation tools rapidly improve and threaten roles in journalism, marketing and advertising.
So what can you do about it?
I can't bang this drum loudly enough: if your job involves research, crunching numbers or creating any kind of written or coded content, you absolutely must embrace AI right now. If you're:
An employee, volunteer immediately to test every AI tool your company considers and become your organisation's AI expert. Prove how you can help colleagues do their jobs faster and better. Also be a really good human i.e. don’t question the wisdom of coming into the office, drive sales and demonstrate your monetary value by thinking of your role from management’s perspective.
A freelancer, proactively approach clients with AI-driven solutions and show them you can offer better, quicker and more cost-effective results by integrating AI into your workflow. Stay ahead of your competitors by mastering AI tools and be prepared to re-learn everything every few months.
A policymaker, stop pretending you understand AI without really using it. Beware of rushing to slash public-sector roles to please budget hawks – unintended consequences are inevitable when you move too fast without understanding the technology. Instead, urgently develop practical policies to retrain workers and manage this transition effectively before widespread disruption hits. The window for proactive response is closing rapidly.
Clearly in some sectors the AI-pocalypse in terms of jobs is already here. Whether you adapt quickly or get subsumed by the waves of AI disruption will depend entirely on your ability to add value above and beyond your output.
AI video of the week
Have we reached the AI hockey stick moment?
I had the privilege of speaking about AI at the Commonwealth Secretariat this week. As often happens with public speaking, my 10-minute slot felt too brief to fully convey just how transformative AI is and will continue to be.
My main messages for the audience of around 60 policymakers were:
Embrace AI immediately: Policymakers need to engage proactively with AI to avoid falling behind.
AI is still in its 'tool' phase: AI is currently a powerful tool, enhancing capabilities and efficiencies, but more revolutionary developments lie ahead.
Energy challenges will intensify: For those managing energy resources, especially in energy-poor nations, AI's expanding capabilities will dramatically increase energy demands, making their roles even more challenging.
Having recently explored cutting-edge releases such as Anthropic's Claude 4 and Google's impressive Veo 3 video generator, I’m convinced we’re at a critical inflection point for AI usage.
Just 18 months ago, many commentators prematurely declared AI's peak hype. As recently as this week, even reputable outlets like The Economist, whose coverage I normally rate, suggested we've hit a "trough of disillusionment."
But this is precisely the divide: between those actively leveraging AI and those neo-luddites who just don’t get AI nor, critically, its value.
This week alone, I created a web extension enabling client feedback on a data dashboard, completed in just a few hours with only a few attempts. Judging by the billions of weekly ChatGPT users and vibe coders, I strongly suspect that my experience is not atypical.
Yes, of course Big Tech is overselling the ‘Agentic Era’ and ‘AI swarms’ and you definitely wouldn’t – or at least shouldn’t – trust any agent with your credit card. However, the lower-hanging fruit of quick and dirty apps and PhD level ‘deep’ research are now within reach of anyone with an internet connection and this early value from AI version 0.1 shouldn’t be underestimated. It's only been six months since agents started appearing on the scene, so give it a year before we can assess their impact. In the interim, experiment, learn and repeat because AI’s value is already here. Just ignore the naysayers who simply can't see it.
Welcome to the token economy
One key message I had to cut from my Commonwealth Secretariat presentation was a profound shift in the way labour and human output is priced. Currently, most of us are paid for our time rather than for our output. However, the AI-first paradigm upends labour economics and shifts pricing towards a token or compute-based economy. A token is a unit of measurement used to quantify AI usage – typically equivalent to roughly three-quarters of a word. For example, if you connect an application to ChatGPT or vibe code with Bolt or Loveable, you will be charged based on how many tokens you use, which includes both the prompt (user input) and the AI-generated output.
For instance, updating 50 documents might involve 100,000 tokens (encompassing both input instructions and AI-generated edits), costing around $2 to $15 depending on the model used. The equivalent human-generated content typically costs between $500 and $2,000, which highlights the stark economic disparity that's reshaping entire industries. Pay pennies for ‘good enough’ or pounds for ‘pretty good’.
At Strategic Agenda, we're now grappling with how to price services such as translation. As clients are sadly becoming more comfortable with AI translation, it calls into question the traditional per-word pricing model in favour of a cost structure reflecting AI usage measured in tokens. This is especially true in the international development sector, which is undergoing savage cost-cutting in the wake of Trumpian economics.
A typical 10,000-word translation project could cost between $1,000 and $2,500 using human translators, whereas an AI-driven translation would cost mere dollars – representing a cost reduction of up to 99%. Now clearly there's a problem with comparing AI translation to professional human translation, but for many clients, especially in these economically uncertain times, ‘good enough’ is increasingly the mantra.
This shift to token-based pricing isn't limited to translation; it will affect any knowledge-based industry that charges per hour or per unit of work. The gig economy is experiencing this disruption first-hand. Content writing, a $6.2 billion annual global market, now faces token-based costs of approximately $3 to $25 per 10,000-word project, sharply contrasting with traditional costs of $500 to $2,500.
Virtual assistance workers in countries like the Philippines, who traditionally charged $5–$15 per hour for administrative tasks, now compete with AI agents costing mere pennies per task. Technical writing specialists in Eastern Europe, accustomed to earning $25–$40 per hour, are finding themselves in direct competition with AI that can produce equivalent documentation for under $50 in tokens.
The traditional labour market, which relied on scarcity-based pricing built around limited human hours, is rapidly transitioning to abundance-based pricing powered by virtually unlimited AI capability. Companies can now hire a research agent for just $10 in tokens, deploy a writing agent for around $5 to produce extensive reports, or engage a coding agent for approximately $20 to develop applications. This marks a fundamental inversion of traditional labour models, from human-driven work with AI assistance to AI-driven processes with human oversight.
It’s going to be a bumpy ride for some time to come, and both service businesses (and employees) need to forget about just charging for their time and think urgently about how they can add value in an AI-first world.
Why Veo 3 changes everything
Google has finally dropped any pretence that its primary objective isn't leveraging your data –from Gmail and search history to Google Suite and, critically, YouTube – to target you with products you probably don’t need. This shift became particularly evident at last week's I/O event, where Google aimed to reassert dominance over rivals who have challenged its market position. Amid the hype surrounding the Open Agentic Web – essentially compact AI models interacting directly within your browser – the standout release was Google's Veo 3, undoubtedly the most impressive video generation tool to date.
What’s so special about Veo 3?
Well, this is the first tool to demonstrate an exceptional grasp of human physics and motion dynamics. This means that it’s almost impossible to distinguish fake content from real. Just take a look at the example above – an AI-generated video of a fictitious car conference in the US, which took just five minutes to generate.
Veo 3 also stands out because it comes with a new video generation suite called Flow. Flow lets you add reference start and end images, then fills in the blanks to solve a major issue in AI video production: ensuring your AI character remains consistent throughout a scene. Google’s vision is straightforward: making video production as effortless as typing or speaking your request.
Currently, Flow is limited to users in the US at a considerable subscription fee of $250 per month. However, those with a Gemini Pro account and a VPN can bypass geographical restrictions to experience it immediately.
What surprised me most, though, is that even in my initial tests, the outcomes were astonishingly impressive. The video below was created by Veo within 60 seconds, based on a single-line prompt describing the scene outside my office. It almost nailed it!
While this technology signifies a substantial leap forward in AI-driven content creation, it also sadly adds another nail in the coffin of creatives worldwide. I don't buy the nonsense of how these tools democratise creativity. In reality, what they do is cheapen the process and also put off generations of budding creatives from entering a creatively challenging but hugely rewarding career.
What we’re reading this week
Anthropic CEO: AI threatens job extinction. It’s worth reading Dario Amodei’s interview in full to get an informed take on the coming jobs “bloodbath”.
UAE is making ChatGPT Plus free for citizens.
UBS, one of Switzerland’s largest banks, is using AI avatars of its analysts to streamline research communications with clients, a move aimed at saving time and engaging customers through their preferred channel.
An AI system tried to prevent itself being shut down by sabotaging shutdown instructions.
Tools we’re playing with
Claude 4: The latest and greatest tool from Anthropic and the model that underpins most vibe-coding tools.
Veo 3: Sadly I only have access to its video generation capabilities via Gemini Pro, but it’s worth playing with to get a sense of where GenAI is heading very rapidly.
That's all for now. Subscribe for the latest innovations and developments with AI.
So you don't miss a thing, follow our Instagram and X pages for more creative content and insights into our work and what we do.