Read More
HK's transport system enters a new phase
22-06-2026 06:00 HKT
2-month-old Danny hospitalised with fever, parents allowed to stay overnight
22-06-2026 04:08 HKT



Specification defines the work. Role Assignment defines the machine’s part in the work.
Most people already do this without naming it. Ask AI to respond as a lawyer and it argues a position; ask it to respond as a judge and it balances both sides. Same facts, different standpoint, different result. This is persona. It works because artificial intelligence is a pattern machine: the persona acts as a lens, aiming the model at one set of patterns, not the default.
Beyond persona, AI can also be assigned a job. For most users, that job is simply to answer, which is only one use of the machine.
The discipline begins when the user chooses that job deliberately. AI can answer, but it can also question, challenge, verify, or orchestrate. Leave the job unstated, and the model falls back to the default: reply.
In Questioner mode, AI is not asked to answer first. It is asked to interrogate the user: what is assumed, what evidence is missing, what the user actually knows. Used well, it does more than improve the prompt; it reveals the shape of the user’s own thinking.
In Challenger mode, AI becomes a hole-poking machine. It stress-tests a theory, attacks weak assumptions, raises counterarguments, and keeps applying pressure until the objections become thin rather than substantive.
AI can also be assigned more procedural roles. In Verifier mode, it traces claims back to evidence and checks whether the output is grounded. In Orchestrator mode, it turns a scattered task into stages, roles, checkpoints, and a sequence. These are only some of the major jobs AI can perform.
Think of AI as a Swiss army knife. The blade is useful, but it is not the whole tool. The discipline is first to recognize that other functions exist, and then to build the habit of unfolding them. The reinvented human does not stop at the blade. They keep pulling out the rest.
Frank Ng is a retired NASDAQ CEO, who co-authors this column with his son Ryan after publishing their book Hey AI, Let’s Talk!