Knowledge got cheap.Judgement didn’t.

Picture a Year 12 student in 2026. Any essay, any assignment, any take-home exam can be produced in under a minute, polished and plausible. The certificate at the end of that education proves a little less every year. Gregor Kennedy, Deputy Vice-Chancellor at the University of Melbourne, pointed the room at the one thing AI can’t do on a student’s behalf: the learning itself.

There is a harder problem underneath. In the foyer afterwards, Michelle Parry-Slater asked the question nobody had a tidy answer for. This cohort lost years of ordinary social learning to COVID. Judgement is learned socially: by reading a room, being corrected, watching how others decide.

For employers, this lands at the front door around 2030. Graduates will arrive with credentials that tell you less than they ever have. Building judgement stops being something universities hand over and becomes L&D’s job from day one.

Experts often know something is wrong before they can say why. That instinct is built over years of repetitions: the junior tasks, the mistakes a mentor caught, the hours spent watching someone better than you do the job. Those junior tasks are exactly what AI is absorbing first. The graduate analyst’s first model. The junior designer’s first draft. The research memo nobody reads twice. One panellist put the problem plainly.

If we automate the bottom of the ladder, we have to rebuild the climb on purpose: simulated reps, scenario practice, structured mentoring, honest feedback. Otherwise, ten years from now, organisations will be full of senior people approving AI output and very few who can tell when it’s wrong.

Defence carried the heaviest version of the afternoon’s question. In a bank, a bad AI decision costs money or reputation. In uniform, it can cost lives. Sensors now see more than any analyst can, and decision windows that used to be measured in minutes are shrinking to seconds. The pressure to let the machine decide is real, and it is only going to grow.

Here is where we landed after the session. The speed of the machine can never become the excuse for removing the human. Accountability for a life-and-death call cannot be handed to an algorithm. But a human in the loop is only a safeguard if that human has the judgement to use it: under pressure, in seconds, with incomplete information. COL Karina Jones was clear that nobody gets there through a single course.

You don’t need to wear a uniform for this to apply. A credit decision. A clinical triage call. A safety stop on a mine site at 3am. Every organisation has moments where “the system said so” is not good enough. Step through the ladder below and decide where your decisions sit.

AI was supposed to give us time back. Ask anyone in that theatre how that’s going. Every tool that saves an hour seems to create two more hours of prompting, checking, reviewing and catching up. One panellist gave it a name: the speed paradox.

When people try to match the pace of the machine, the first thing to go is the slow, deliberate thinking that judgement depends on. Decisions get made at the speed of the tool instead of the speed of understanding. The organisations that win this decade won’t be the fastest. They’ll be the ones whose people know when to slow down.

Every thread ended in the same place. Emma Weber took it furthest, into belief, meaning and what it takes to bring your whole self to work alongside AI. Brad Coumbe, who leads learning at CommBank, walked out holding on to her point that it is passionate people who make AI into what we want it to be. In the foyer afterwards, Rod put the whole afternoon into a single line.