I used to hear “human plus AI” and immediately think, yes, that sounds sensible. Very balanced. Very modern. Very likely to appear on a consulting slide with two overlapping circles and a word like “synergy” in the middle.
Then I read more of the research, and the annoying truth is that human-AI collaboration is not automatically better. A meta-analysis by Vaccaro and colleagues looked across more than 100 experiments and found that human-AI combinations often beat humans alone, but do not always beat the stronger standalone partner. In plain English: adding a human to AI, or AI to a human, does not magically create brilliance. Sometimes it creates a meeting.
The part I find most useful is that collaboration depends on the task. AI can be very helpful for content creation, drafting, summarising, and routine support. But for decision tasks, especially where the AI is already strong or the human role is unclear, the combination can become messy. The human may over-trust the system, under-trust it, or approve something without really understanding it.
So the question is not, “Should we put a human in the loop?” The better question is, “What is the human supposed to do in the loop?”
Are they checking facts? Adding context? Catching ethical risks? Improving tone? Making the final decision? Or are they just there so the organisation can say there was oversight?
That last version worries me. It gives everyone comfort without giving anyone power.
My reflection is this: human-AI teams need instructions, not optimism. The best collaboration is designed around roles, timing, disagreement, and accountability. Otherwise, we are not building a smarter team. We are just creating a workflow where both the human and the machine can be confused at the same time.