Let's start with the elephant in the room that every LinkedIn thought leader conveniently ignores while posting their AI-generated infographic about "the future of work": the productivity gains from AI do not accrue to the worker. They accrue to the business that no longer needs as many workers.
We have been here before. Automation in manufacturing. Spreadsheets eliminating accounting pools. ATMs replacing bank tellers. And every single time, the optimists won the narrative — "technology creates more jobs than it destroys!" — right up until the specific job you held vanished.
This time is different. Not because AI is magic. Because for the first time, automation is eating cognitive work, not physical work. And cognitive work was supposed to be our moat.
"You are not using AI as a tool. You are using it as a mirror — showing the system exactly how you think, so it can think without you." The delegation trap Here is what nobody will say at your next AI strategy workshop: every time you delegate a task to an AI — writing a report, summarising a meeting, debugging code, drafting a proposal — you are removing yourself from the learning loop. You stop practising. Your judgment stops sharpening. The model gets better. You get worse.
Junior consultants who used to learn by writing first drafts now receive polished AI outputs to "review." Junior engineers who used to grow by wrestling with bugs now ask Claude to fix them. The training pipeline for the next generation of expert human judgment is quietly being dismantled — and we're calling it "upskilling."
A question worth sitting with: If you lost access to every AI tool tomorrow, would you be more capable or less capable than you were two years ago? If your honest answer is "less capable" — you have already begun your own replacement. The productivity illusion Businesses are reporting massive productivity gains. Outputs are up. Time-per-task is down. Headcount is flat or shrinking. Investors are thrilled. This is being called a transformation. It is also a compression — squeezing the same economic output from fewer humans, with the difference going not to wages but to margins.
The uncomfortable data point: in every industry where AI productivity tools have been widely adopted, the ratio of economic value created per human employed has increased significantly — while median real wages in those sectors have stayed flat or declined. The productivity miracle is real. The shared prosperity narrative is fiction.
What "AI-augmented" actually means at scale When a company says it is "AI-augmenting" its workforce, it means one of three things. First: the same output from fewer people (layoffs, just announced later). Second: higher output from the same people, creating pressure to reduce headcount in the next planning cycle when leadership asks why they can't maintain this output with 20% fewer people. Third: genuinely new value creation that wouldn't have existed otherwise — rare, celebrated, and the exception, not the rule.
The third scenario is the one in every keynote. The first two are what's actually happening in most quarterly planning meetings right now.
"The companies celebrating AI adoption the loudest are often the same ones quietly initiating 'workforce restructuring' plans. Read both press releases together." So what do you actually do? I am not arguing you should avoid AI tools. That ship has sailed, and using them is now a baseline professional competency. What I am arguing is that you should be intentional about which parts of your thinking you outsource — and which parts you deliberately protect.
Use AI to compress execution time. Never use it to replace judgment formation. Use it to explore options faster. Never let it make the decision. Use it to handle the mechanical. Refuse to let it handle the meaningful.
The professionals who will be genuinely irreplaceable in five years are not those who are best at prompting AI. They are those who are best at knowing when not to — and who have continued developing the domain depth, the contextual judgment, and the interpersonal fluency that no language model can replicate, because those skills emerged from thousands of hours of doing the hard work themselves.
The real controversy Here is the take that will get me unfollowed: the AI safety debate has been almost entirely captured by the wrong concern. We are arguing about superintelligence, misalignment, and robots taking over the world — while the more immediate, more certain, more tractable catastrophe unfolds in slow motion across every knowledge economy on earth.
Not AGI destroying humanity. Automation destroying the middle-class pathway that took a century to build, happening fast enough to cause enormous societal disruption and slow enough that no single news cycle captures the full picture.
The danger is not that AI becomes too smart. The danger is that we let it make us too comfortable with being less capable — and that we only realise it when the leverage has already shifted permanently.
I work at the intersection of AI adoption and cloud infrastructure daily. These views are my own and they are deliberately uncomfortable — because comfortable views don't force the conversations that matter.