I recently saw someone on X claim that companies are already laying off anyone who can’t use AI.
So I asked for the source.
Crickets.
That’s the AI conversation about work in a nutshell. One camp is sure AI is about to wipe out most jobs. The other talks as if we’re a few breakthroughs away from machines doing everything while the rest of us drink piña coladas.
I don’t find either story useful.
Yes, some companies are cutting jobs while talking about AI. Plenty of them also over-hired during the cheap-money era, want better margins, or are restructuring for reasons that have little to do with AI. Anyone who claims they can cleanly separate those causes from a press release is selling something.
So I went looking for data.
In August, the Bank of Canada published an analysis of how occupations with different levels of AI exposure are faring. Workers in highly exposed occupations aren’t losing or leaving their jobs any faster than everyone else.
The change is in getting hired.
Between 2015 and 2019, unemployed people looking for work in the most AI-exposed occupations were 2.2 percentage points less likely to find a job than those in the least exposed. By 2025, that gap was 13.9 points.
The gap in job separations barely moved.
The authors call it an early signal, not proof that AI caused the shift. The labour market has been weird for plenty of other reasons. Fair enough. I’d still pay attention.
If you’re employed, it’s tempting to read this as somebody else’s problem. Most of the pain is landing on job seekers, and a lot of them are young.
But think about your exit plan.
If you’re quietly planning to leave your job for a similar role somewhere better, that’s the door that’s narrowing. Employers are slowly hiring fewer people to do the old version of your job.
Your job is changing either way. Better to redesign it yourself while you still have room to experiment.
Pay attention to the tasks inside your job
You can’t cut your way to growth, and the good organizations know it.
That said, a lot of organizations got bloated over the years. They added layers of process, reporting, coordination and admin work that create very little value.
Your job has some of that too.
Reports nobody reads. Meetings that lead nowhere. Reformatting information from one system into another. Hunting through documents. First drafts that restate what everyone already knows.
That’s where your risk sits. The more of your week goes to this stuff, the more exposed you are. Fixing it is on you.
If software can do part of a task, stop defending the task. Hand that part to the tools and keep ownership of the result. Review the output, apply your judgment, and follow your organization’s rules on data, privacy and security. You’re still accountable. You’re just producing more.
Where you can, stop doing useless shit altogether.
Decide where the recovered time goes
I keep seeing people talk about using AI to get a 40-hour job done in 20 hours.
I get the appeal. But if your contribution visibly takes half the effort, you’ve handed your employer a good argument for needing less of you.
The opposite move has its own trap. Pour every recovered hour back into your job and you’ve built a faster treadmill. More output, same cage.
Treat recovered time as capital, and decide where it goes.
Some of it goes into making your job harder to replace. The bottlenecks everyone else has learned to tolerate. The broken processes nobody owns. The decisions that need judgment and context.
Some of it goes into your Work. The thing you’d do whether or not anyone paid you, and the skills, relationships and projects that move you toward it.
The split is your call.
Learn by rebuilding real work
People make this harder than it needs to be by deciding they have to “learn AI.” They sign up for courses, watch tutorials and play with random prompts.
You’ll pick up a bit that way. The faster route is to pick a real piece of your work and rebuild how you do it.
I’ve done this with this newsletter. Research, editing, production assets, the metadata in my content system. Each piece started as a clumsy chat and is now a workflow with its own instructions, reference material and examples. The first outputs were rough. Some were bad. Every time, I had a choice: fix it myself and move on, or teach the system what I wanted.
Here’s how to start.
Choose something recurring that eats time and doesn’t need your full judgment at every step.
Explain the work to the AI. Give it the inputs, the output you want, the constraints and examples of what good looks like. Ask where it can help. Break the task into parts and start with one.
It can research, sort or classify information, write a first draft, check a result against a standard, or prep material for the next step.
The first attempts will be bad.
That’s where most people make the wrong move. They take the bad output, fix it themselves, and carry on. The task gets done, and the system learns nothing.
Show it what was wrong. Tell it what you wanted instead. Give it a better example. Update the instructions or the reference material. If the workflow is worth keeping, build the correction in so you never explain the same thing twice.
Over time, a one-off chat becomes a repeatable workflow with instructions, source material and feedback built in.
Then move to the next task.
You’ll produce more without working more. You’ll also start seeing work as a system you can redesign. That skill goes wherever you go, including into your own Work.
Your expertise is the part worth protecting
The Bank of Canada found AI exposure was highest in routine information work and lowest in work that depends on judgment, physical interaction or specialized human skills.
That matches what I saw in AI governance work in government.
AI is very good at chewing through information and producing competent output when the task is clear.
What it doesn’t have is your accumulated context.
It doesn’t know why the last three attempts failed. It doesn’t know which stakeholder will object, which technically correct answer will never survive contact with your organization, or which shortcut will create a mess six months from now.
You know those things because you’ve spent years doing the work. That’s the asset.
An experienced professional who knows the domain and can redesign the work with AI is in a very different position from someone doing the same tasks the same way, hoping the tools never get good enough to matter.
Ignore people who can’t bring data
You can find someone online to back almost any AI story you want. Most of them can’t produce a source either.
The data so far is less dramatic. In a Statistics Canada study cited by the Bank of Canada, almost 90% of businesses that adopted AI reported no effect on staffing levels. Adoption is uneven. Some tasks are disappearing. Others are becoming more valuable.
Nobody knows where all of this ends. Watch how your own work is changing.
Find the routine parts of your job and spend less of your time on them. Then decide where the recovered time goes.
If hiring is shifting before layoffs do, I’d rather adapt while I still have room to experiment than wait for the market to make the decision for me.
Sources
Bank of Canada: Early signs of AI-driven adjustments in Canada’s labour market