I've run over 40 training sessions on AI in the past couple of years. Same room, different organisations. Same question every time, buried under different phrasing: Is this coming for my job?

The honest answer isn't reassuring or simple. It's granular. AI doesn't replace roles. It replaces tasks. Your role probably contains all three of these categories, and your job now is figuring out which is which.

What AI replaces

Some work is purely productive machinery: get from here to there, turn the input into the output, repeat.

First-draft writing. Routine research and synthesis. Data wrangling. Building campaign briefs from templates. Scheduling admin. Basic creative production — stock imagery, simple graphics, template variation. I spent roughly 40% of my week on this category five years ago. Now it's about 5%. Not because I've stopped working. Because I'm not retyping anymore.

I used to move information from my brain into a document in the same form it was already in. Raw synthesis — take three sources, format them, write a covering line. Machinery. AI does that now. Does it do it better? Not always. Does it do it fast enough that I'm not the bottleneck anymore? Yes, reliably.

The liberation is real. You don't notice it until it's gone — that fatigue of doing the same shape of work repeatedly, the context-switching between thinking and typing, the cognitive waste of transcription.

Where you need to be careful: not all "writing" fits here. Strategic narrative doesn't. A board paper that needs to land a specific way doesn't. A performance review that names why someone mattered to the team doesn't. These aren't first drafts dressed up. They're judgement calls with stakes. AI can help you get there faster — quicker clarity on structure, sharper phrasing options. But the choice about what matters is still yours.

What AI extends

This is where work gets interesting. AI doesn't replace your skill here. It expands it.

I've spent fifteen years building bespoke models. Scoring systems. Assessment tools. Audit methodologies. Frameworks that fit a problem nobody else has quite solved the same way. That work required time: weeks in spreadsheets, months of prototyping, iterating, testing against reality.

Now I can prototype in days. Build, test, throw away, rebuild. The bottleneck shifted. It's not "can I build this?" anymore. It's "should I build this, and what am I trying to solve?"

AI doesn't replace my judgement about what matters. It compresses the iteration cycle so I can make better judgements. I can sketch five versions of a model in the time it used to take me to finish one. That means I'm not locked into my first idea. I can explore the problem space properly.

The same applies to stretching into adjacent disciplines. I don't have a background in conversion rate optimisation or data architecture or advanced statistical design. But I can learn fast enough now to work productively across those boundaries. I have a baseline from AI, then I shape it into what the problem actually needs. Five years ago, that would've meant hiring someone or spending months on foundational reading.

But notice what's constant across all of this: you are still the filter. You're still choosing what to build, what to keep, what to test, what to discard. The range has extended. The judgement is still yours.

What AI can't touch

This is the category I think people get wrong most often.

AI can't train a room full of people who don't believe in the change yet. It can't walk into a tense strategic meeting and help a leadership team make a hard call about what to double down on. It can't diagnose why the sales team isn't adopting the new process. It can't build the trust that makes someone willing to do something difficult in a different way.

That work is human. It's relational. It depends on presence, on reading a room, on knowing when to push and when to back off, on being credible enough that people will follow you into uncomfortable space.

I've trained hundreds of people on this stuff. I couldn't have done it from a script. Couldn't have done it without being in the room, watching what landed and what didn't, adjusting on the fly, answering the real question underneath the question someone asked.

The counterintuitive bit: as AI handles more of the production work, this becomes more valuable, not less. When the bottleneck was "who can write the brief and build the deck and schedule the calls," you needed a lot of execution-level people. Now you need fewer people who are faster at production. But the leverage point shifts to the person who decides what to produce and who can bring people along while it's changing.

Facilitation matters more when production speeds up. Strategy matters more when execution becomes cheaper. The human who can hold a vision and build conviction becomes rarer and more useful.

Making it useful

Look at your own week. Split it into those three buckets. Don't be romantic about it. If you're spending time on something that's purely throughput — moving information from state A to state B with no judgement baked in — that's a candidate for replacement. You probably won't miss it, and you'll free up space.

The work that extends your skill is where you lean in. Find the edges you've been avoiding because they required too much setup. Sketch more models, explore more adjacent disciplines, build your knowledge systems faster. The time you save in bucket one funds the ambition you can now afford in bucket two.

And the work that AI can't touch? Protect it. Don't let it get crowded out. The more your organisation relies on AI for production, the more they need you for conviction. That's not a threat. That's your actual leverage point.

The question isn't whether AI is coming for your job. It's already here. The question is whether you're using it to do the same job faster, or to do a different job altogether.