This is where I meet the mirror most, and where most of us do — at work. And it is where the line between “AI made me a superhero” and “AI made me a replaceable operator” is thinnest. The frightening part, which I have felt in my own hands, is that you cross it without noticing. The cape and the strings look the same until someone asks you to explain the thing you just handed them.

Start with which tasks, not which tools

I keep the hardest fact from the first series taped above my desk: human-AI teams show positive synergy on creation and problem-framing, and negative synergy on judgment and decisions — sometimes performing worse than the AI alone. That is my operating manual at work. Lean in where the machine genuinely lifts me: drafting, exploring options, boilerplate, first passes, summarising, brainstorming against a wall. Slow down and stay the human where it quietly hurts me: the decision, the evaluation, the judgment call with real stakes. The professional's skill is no longer just doing the task — it is knowing, task by task, which mode you are in. I get it wrong when I stop asking.

The jagged frontier is your job now

In the workplace, the “jagged frontier” from the first series is not an abstraction — it is the thing I am paid to police. AI is brilliant, then confidently and invisibly wrong, and the boundary between the two is not marked. Being an expert used to mean doing the work. Increasingly it means being the one person in the room who can tell where the machine's competence ends in your specific domain, and who catches the plausible error before it ships. “Human-in-the-loop” is a comforting phrase that is only worth anything if the human in the loop still has the judgment to overrule the machine. A supervisor who cannot evaluate the work is not supervising; they are rubber-stamping — and I have caught myself, on a tired afternoon, one click away from exactly that.

The paradox that should genuinely worry you

Here is the trap the 2026 research names outright, the oversight paradox: AI use is concentrated in exactly the expert, judgment-heavy knowledge work — analysis, code, technical writing — where meaningful oversight matters most. But using the AI erodes the very expertise you need to oversee it. Cognitive offloading, automation bias, less practice, missing feedback loops — they quietly hollow out the judgment that was your whole value. Tellingly, the workers who use AI most cautiously are the high-aptitude ones; treating the machine as authoritative is a low-skill tell. So the discipline I hold myself to is concrete: keep doing some hard work unaided, deliberately, to keep the muscle that lets me catch the machine. Gartner expects the atrophy to be severe enough that half of organisations will require “AI-free” skills assessments — the labour market is about to start checking whether you can still think without it.

The puppet trap

Now the failure mode in one image. If I become a pure operator — pasting prompts in, shipping outputs I cannot evaluate out — I have made myself the single most replaceable link in the chain: the interchangeable hand between the prompt and the paste, adding no judgment the next person (or the next model) couldn't add. That is the puppet. The way out is the “human premium”: as the machine commoditises the production, value migrates up — to judgment, taste, knowing which problem is worth solving, and the client trust that no model can hold. Do not defend the task; you will lose. Move up to the judgment, where you win. That is the move I bet my own work on.

Two more workplace disciplines

Make the mirror argue with you. An AI that flatters your analysis (the sycophancy of the first series) just produces confident nonsense faster, with your name on it. I explicitly ask it to attack my reasoning, find the hole, argue the opposing case. A yes-man is a liability in a suit.

Mind your sovereignty. My firm's and my clients' crown jewels — the strategy, the case file, the code, the relationships — do not belong, unexamined, in a rented mirror I do not control. I make it my business to know where my professional self is stored, and on whose terms.

And the honest balance

None of this is an argument to refuse the tool — that is the other way to lose, and I have watched people lose that way too. The cliché is a cliché because it is true: AI will not take your job, but a colleague who wields AI with judgment will take it from a colleague who won't. And the deeper asset is not any skill you have today; the World Economic Forum expects 39% of core skills to change by 2030, so I treat re-skilling itself as the job. The mature professional is neither the luddite nor the operator. It is the disciplined hybrid: augmenting aggressively, keeping judgment sharp, making the machine disagree, and learning without stopping.

So — wear the cape. It is real, and the person beside you who refuses it will fall behind. But the moment you can no longer evaluate what the machine hands you, the cape has quietly become strings, and you have become the thing being worked. Stay the judgment. In the age of the augmented professional, the judgment is the job — and keeping mine is a fight I have to win again every week.