What has already gone
Manual bidding went years ago and almost nobody misses it. Since then: budget pacing, negative keyword harvesting, ad rotation, dayparting, anomaly alerts, most reporting, and the first draft of nearly all ad copy.
None of that was the interesting part of the job. It was the part that filled the week.
What has not gone
Deciding what a customer is worth. Margin, lifetime value and which customers you actually want more of are business facts that live outside every ad platform. Nothing in the account can derive them, and everything in the account optimises towards them once told.
Deciding what to buy. Which markets, which products, which moment. A system optimises within a plan. It does not have one.
Noticing the plan is wrong. The hardest judgement in the job is telling the difference between a bad month and a broken assumption. Automation is good at the first and structurally poor at the second, because both look like numbers going down.
Arguing with the business. Telling a founder their target ROAS is incompatible with their growth target is not an optimisation problem. Somebody has to be in the room.
The job that is left
Fewer people, doing more senior work, across more channels each. The person who used to run one platform well now sets the objectives for several and spends their time on the things that are actually decisions.
The scarce skill stops being platform expertise and becomes knowing what the business is trying to do and being able to express it as something a system can optimise towards. We argued this at length when AI made execution cheap and strategy scarce: the advantage moved to the people who can see how everything connects.
If you are a media buyer now
Concretely, and in rough order of return:
- Learn the business, not the platform. Margin, contribution, payback period, the difference between revenue and profit at the SKU level. This is the input every system needs and almost no one supplies well.
- Get good at measurement. Incrementality, attribution limits, why two platforms disagree. Automation makes measurement quality decisive, because every automated decision inherits it.
- Be the person who sets guardrails. Deciding what a system may do unattended is a new job and it is not going to a junior.
- Stop competing on execution speed. That race is lost. It was lost before the current wave.
The honest risk
The optimistic version of this answer says technology creates more jobs than it destroys, and over a long enough horizon that has usually been true. Over the next few years, in this specific job, the more likely shape is fewer roles, more senior, and a middle that thins out.
The real cost is the training ground. Junior media buyers learned judgement by doing the tedious work: pulling search term reports, watching what happened after a change, building the pattern recognition that senior judgement is made of. Automate all of that and the ladder loses its bottom rungs, without anyone deciding to remove them.
Nobody has a good answer to that yet. Anyone who tells you the transition is painless is selling something, and so are the people telling you the job disappears entirely. What actually changes is which parts of the work a system is allowed to do on its own, and that line is drawn by a person.