When to hire a human and when to hand it to AI

AI can absorb tasks. It cannot own the consequences.

When half my content team left, my first instinct was to replace them and rebuild what had disappeared. Then I paused. By that point, AI could already handle much of the work those roles had been designed around. Research came back in minutes. Rough drafts arrived quickly. Basic analysis no longer clogged the week.

Hiring the same team again would have meant rebuilding around work that had already changed.

The usual debate asks whether AI can replace a person. That makes for good conference panels and terrible management decisions. A job is a bundle of tasks, and those tasks do not change at the same speed. AI may handle the research while a person decides what matters. It may draft the proposal while someone else takes responsibility for the promise inside it.

The useful question is where judgement enters the work.

AI is most useful when the task is clear, the output is easy to inspect and a mistake can be reversed before it reaches a customer. Summaries, research briefs, first drafts, routine analysis and content variations fit that description. A capable person can give the system direction, review the result and move much faster than before. That does not make the person optional. It changes what you are paying them to do.

The value moves away from producing every word and towards deciding which words deserve to exist. The human frames the problem, provides context, rejects the plausible answer and notices when the work is technically correct but commercially useless. AI makes production cheaper. It makes weak judgement easier to distribute too.

The line becomes clearer when the work carries consequences. A model can help prepare an enterprise proposal. It cannot sit in the meeting and notice that the procurement objection is really fear about implementation. It can generate a campaign. It cannot own the reputational cost when the message lands badly. It can recommend a decision. It does not have to explain that decision to the board six months later.

Use AI heavily where the work is reversible and reviewable. Hire a person where the company needs context, trust, accountability and decisions under uncertainty.

There is another part of this argument that founders tend to avoid. Junior roles contain plenty of repetitive work, which makes them obvious targets for automation. They are also where people learn how good work is made. The first drafts, basic analyses and customer calls that look inefficient from a spreadsheet are often the apprenticeship that produces future judgement.

Remove every junior task and the economics look excellent for a while. A few years later, the company may wonder why nobody has the experience required to become senior.

Keeping busywork for sentimental reasons solves nothing. Junior roles should change. Give people AI from the start, then move their learning closer to the customer and the decision. Let them review outputs, test assumptions, sit in difficult meetings and understand why one answer works while another merely sounds polished. A junior employee using AI well can become useful faster. AI operating without anyone learning from the work leaves the company with output and no bench.

Before opening a role, break it into its actual tasks. Which parts are predictable? Which mistakes are cheap? Where does the work depend on private company context? Who will judge the output? Who owns the result when it goes wrong?

That last question matters most. When nobody inside the company can tell whether the answer is good, AI is not saving money. It is producing unsupervised risk at impressive speed.

Sometimes the answer will still be a full time hire. The function needs daily ownership, relationships must be built, decisions arrive constantly and someone has to live with the consequences. Sometimes a smaller team with better tools will do more than the old structure. Sometimes the work should disappear rather than be automated or rehired.

When my team changed, I was not choosing between people and software. I was deciding which work still needed to exist, which parts could be handed over and where judgement had become more valuable.

AI will reduce the number of people required for some tasks. It will also make it easier to hollow out the path through which people learn. The mistake is counting the salaries removed without asking who will own the difficult decisions and who is learning to make them next.

That is the hiring decision AI cannot make for you.

See you out there.

Martin