Last Wednesday, Australian logistics software company WiseTech Global announced it would cut 2,000 jobs as part of its AI transformation.
This was the first time that mass AI-related job losses had been announced in Australia.
And on Friday, Jack Dorsey, CEO of ASX-listed Block, which acquired Afterpay for $39 billion in 2022, announced that about 4,000 of his 10,000 employees would be made redundant.
“We’re not making this decision because we’re in trouble. Our business is strong,” he said.
“But something has changed. We’re already seeing the intelligence tools we’re creating and using, combined with smaller, flatter teams, enabling new ways of working that fundamentally change what it means to start and run a company. And it’s rapidly accelerating.”
Last week, Paul Tyler, CEO of internet provider Superloop, said almost all customer interactions are now done through AI. “Our philosophy is that if you have to call us, we did something wrong.” (Humans still answer Superloop phones.)
Mr. Tyler said Superloop has not cut any jobs and achieved this by increasing production without hiring people through natural attrition.
What are these companies doing with AI? How will they use it and how far will it spread?
Companies are now able to go beyond just asking AI questions and “hiring” AI agents, also known as agent AI, to work for them. (Unsplash: Aydin Guerinrekab)
Agent AI is being used to replace human workers
We are in the midst of a fundamental shift in the way we use AI, and it is happening much faster than most of us realize. There is currently a huge gap in understanding between those working in AI and the rest of the world.
This year began not only with incredible advances in both AI and robotics, but also with intensely polarized debates about whether the future will be a dark dystopia or a bright and prosperous future.
Regarding the last two points, I don’t have much to offer. Technology is way above my pay grade, and as Yogi Berra wisely said, “The future is hard to predict, especially about the future.”
But Wisetech, Superloop and other Australian companies I spoke to give us a window into how AI is replacing human workers and how it is changing the world.
Companies are now able to go beyond just asking AI questions and “hiring” AI agents, also known as agent AI, to work for them.
Agents are software designed to apply AI reasoning and data to specific business tasks, such as responding to customer inquiries, developing marketing plans, and processing legal documents.
These can be purchased off-the-shelf or, as in the case of Superloop, built by in-house engineers customized specifically for the company.
They sit on top of large-scale language models (LLMs) such as ChatGPT, Gemini, Claude, and Grok, from which they purchase “intelligence.”
A rough analogy is that the LLM is the brain, and the agents are the nerves, bones, and muscles that make the brain’s thoughts work.
As with manufacturing, China is encroaching on the U.S. AI model. (Reuters: Dado Lubitsch)
The cost of collapsing intelligence
Every company I speak to says they are switching LLMs and buying the cheapest tokens they can find.
Tokens are the unit of AI “thought” (the currency of machine intelligence), and 1,000 tokens are equivalent to approximately 750 words.
An agent attempting to solve a complex problem repeatedly pings the LLM, with each ping consuming an “input token” (instruction), followed by an “output token” (decision or documented product).
Token prices have plummeted in recent years. Each LLM has a slightly different price, with prices falling unevenly depending on the complexity of the task.
The price of medium intelligence (10-20 on the Artificial Analytical Intelligence Quotient) has fallen from $30 USD to 3 cents in the past two years.
The fall in prices was caused by competition. In 2022 and 2023, OpenAI’s ChatGPT dominated the field and tripled in price with a huge increase in IQ, but then Anthropic launched Claude, Google came up with Gemini, and the price dropped in three years.
One result of this is that LLM tokens have become a commodity and the companies that manufacture them are rapidly becoming low-margin utilities. This is not how the stock market is pricing at this stage and leads to predictions that the bubble will burst.
Of course, similar to power companies that sell electricity as a retail product to the public and wholesale to industrial customers, LLM not only sells subscriptions to the public, but also wholesales tokens to businesses and is starting to generate significant revenue from it.
All have free tiers, some have entry-level prices of around $10/month, but all have standard/premium prices of $15-20/month.
There are also retail “power user” subscriptions for $200 (for Claude, ChatGPT, Perplexity), $250 (Gemini), or $300 (Grok) per month, which are the “agent” versions of the AI. In other words, they act like the agents that companies use to replace human workers, so they do the work for you instead of just telling you the answers.
But as with manufacturing, China is squeezing them.
Experts say China’s DeepSeek is not as good as the U.S. model, but it’s usually good enough, and by offering the full range of PhD-level inference for free, it’s putting a lot of pressure on industry pricing.
This threatens to eliminate the standard $20 tier for Americans and force them to focus on agent versions.
For both developers and enterprises, DeepSeek is pay-as-you-go, per token. The price is USD 28 per million tokens for “Chat” and “Reasoner” levels and USD 2 per million tokens for DeepSeek-V3.2-Exp.
When I asked DeepSeek’s chatbot to explain the differences between the two, Chat and Reasoner both said something confusing when using DeepSeek-V3.2-Exp.
While chat is the standard mode of answering questions immediately, Reasoner is a “deep think” mode that generates internal thought chains, “allowing you to break down complex problems, double-check your logic, and reach more accurate conclusions.”
I looked at all of this in some detail to show that the entire retail and wholesale ecosystem was born from scratch in about a year in a way that most people still don’t understand.
Loading
“Sleeping next to a tiger”
What makes this product different is that LLM Utilities is selling a product that is becoming increasingly commoditized: a product called intelligence.
It’s called artificial intelligence to differentiate it from the human kind, so it doesn’t include empathy or other human emotions (yet), but it’s basically the same thing.
As Sam Altman, CEO of ChatGPT developer OpenAI, recently pointed out when defending the use of AI resources, human intelligence has a long and complex background.
“Not only that, but creating humans required the extensive evolution of the 100 billion humans who have lived, learned how to avoid being eaten by predators, learned how to understand science, and so on.”
And now, as economic blogger Noah Smith recently pointed out, we may no longer be the most intelligent beings on the planet.
“It’s not just that AI could take your job, put millions of people on welfare, or give us endless amounts of free software, etc. It’s that for the first time in recorded history, humans are no longer, or soon will be, the most intelligent beings on the planet in any meaningful, functional sense of the word.
“We’ll be sleeping next to tigers for the rest of our lives.”
Alan Kohler is a financial presenter and columnist for ABC News and a contributor to Intelligent Investor.
