The most 'AI-exposed' jobs keep winning: how quickly is AI really coming for your job?
From looms to low-code, history shows that automation kills tasks long before it kills the job.
In 1804, Joseph Marie Jacquard patented the Jacquard loom, a machine that could automate intricate weaving patterns.
Overnight, the activity of manually lifting and lowering threads vanished, but the job of a master weaver did not disappear immediately, because weaving was more than pulling levers.
Weaving meant also sourcing the right materials, designing patterns, maintaining relationships with buyers, training apprentices, ensuring the product fit the market, etc etc
Factories learned quickly: if you only replace the activity and keep the rest, you still need humans to coordinate, troubleshoot, and even innovate. Once the entire system around that activity was redesigned (eg, supply chains and markets) the job was redefined, not just the activity. That is when real displacement happened.
Same story with AI today.
People think: “If AI can do the writing, it can replace the writer.”
Maybe, but a writer is not a collection of typing sessions.
It is research, context-setting, editing, knowing what NOT to write, building trust with an audience, shaping ideas to fit a purpose.
Replace one part and you get a tool.
Replace the whole system and you get a “paradigm” shift (I dislike the term but I keep using it, damn)
That’s why the replace activities, replace jobs logic is seductive but false: it skips the phase where you actually have to re-engineer the whole system to make that replacement real.
Most revolutions die in that gap, actually.
In fact, breaking work into atomized tasks and assuming you can simply replace them with AI ignores the emergent properties of a job.
The value of a job is so much more than the sum of activities, just like a football team is not simply eleven players on a pitch.
The interactions, sequencing, tacit knowledge, context-switching, informal communication, and decision-making glue together the whole.
Strip them into fragments, and you kill the system that makes those fragments valuable in the first place.
That is why sometimes, when you extract a top scorer from a team and put him into a different team, the end result is a lemon...
🚨 Get my book Beyond Slides: Stand out, become irreplaceable, and win big in consulting
If you want to develop the part of your consulting toolkit that no AI or template can replace (eg, the ability to influence decisions, win trust in the room, and navigate complex situations) Beyond Slides was written for you.
A #1 Amazon Best Seller in the Consulting category in USA, UK, Australia, and Italy, Beyond Slides is out in e-book and paperback format on Amazon (also available as an audiobook in the USA) and whatever you buy books.
Thank you for your support.
By the way, we have seen this play out before (as usual, nothing new under the sun).
In the crypto/web3 DAO (Decentralized Autonomous Organization) hype, people thought that if you paid individuals for micro-tasks, you would replace the need for organizations.
What happened instead?
Governance problems, lack of accountability, coordination costs through the roof. The DAO model failed, but not because the idea of paying for output was wrong... It failed because it treated work as modular Lego bricks rather than as a living organism.
The same happened in the early “gig economy” wave.
Platforms like Mechanical Turk proved you could atomize certain jobs into microwork. But the more complex the job, the more those small tasks needed to be embedded in context: something no platform could fully capture without rebuilding the very structure they were trying to replace.
A clean modern tech parallel is the “low-code/no-code” boom of the late 2010s.
When Bubble, Airtable, Zapier, and about 225 other platforms started trending, the narrative was: “Folks, you no longer need developers!! Just drag, drop, and automate your way to a product!!!”
And yes, certain activities of software engineering vanished for small, contained use cases, eg prototyping a form, building a database front-end, automating a workflow, etc
Of course, this did not kill the job of a developer, because building software is more than slinging code snippets (ie, the job includes translating a vague problem into a concrete architecture, ensuring security, scalability, and compliance, integrating with messy, real-world systems, managing trade-offs between speed, cost, and quality, coordinating with product, design, and stakeholders who change their minds, and probably other 10 things I’m forgetting).
You can replace the type code activity, but unless you also re-engineer everything around it (at a minimum, how companies scope projects, maintain systems, manage change) you still need the role.
As per Jevons paradox (yes, him again!), low-code/no-code expanded the demand for engineers. Once more non-technical teams could prototype quickly, more projects got greenlit. And the very engineers we were told would be obsolete had to come in to productionize, integrate, and maintain them!
I believe that is exactly the trap with AI predictions today.
People point to the automation of activities and think they have seen the end of the job. In reality, they are just front-row to the start of a system redesign that might take years or decades, and might create more demand for the role before it destroys it.
To wrap it up, just look at this chart:
If ‘AI exposure’ really meant ‘AI replacement,’ this chart would be flipped upside down.
The reality is, instead, that the most AI-exposed roles have lower unemployment, which is exactly what we saw with the Jacquard loom, the gig economy, and no-code: automate the task, and the surrounding work often expands.
Until you re-engineer the whole system, the job is stubbornly alive. The hard-to-automate layer is not the individual keystrokes or prompts, but the orchestration.
Did I make sense? Let me know in the comments…
👋
👀 Links of interest
A few corners of the internet you may find interesting:
I recently joined Nik Hulewski on his popular podcast Nikonomics to talk many things consulting. I hope you enjoy the episode: it was a fun half an hour... and, at some point, you can even hear Nik sing the Backstreet Boys (that alone makes it worth it!)
The Leaders Toolkit is a deck of 52 tools, frameworks and mental models to make you a better leader (use code CONSULTANT10 for 10% off);
The Consulting Intel private Discord group with 230+ global members is where consultants meet to discuss and support each other (it’s free).
“Just as the internet forced every bank to go digital, and mobile forced every bank to go omnichannel, a new interface (that does not involve a screen, an app, or a human hand) is emerging: the AI agent” - A recent LinkedIn post of mine if you are into the intersection of AI and Banking…
This post is exactly right. AI replaces activities not jobs - same as every previous technology. The paradigm of "which jobs will AI replace" is actually very limiting. AI will do things differently. In some cases it will just like for like replace tasks, in others the new capability will result in a whole process being re-engineered with completely new sets of activites.