The shift Mollick names: from co-intelligence, where you correct and steer the AI mid-task, to conjuring, where one vague prompt returns a finished artifact and there are no decision points left to intervene in.
GPT-5 Pro spent under ten minutes critiquing Mollick's own job market paper, ran Monte Carlo checks in code, reinterpreted his fixed effects, and surfaced a small inconsistency between two tables that a year of expert human review had missed.
Reinforcement-learned agents pick their own steps, so nobody selects the plan, not even the builder. The step lists these tools display are summaries of the work rather than the work, and some tools show essentially nothing.
Verification cost is the hidden variable. Checking a wizard's output is sometimes far cheaper than doing the work yourself and sometimes far more expensive, and some outputs are sophisticated enough that you could not check them if you tried.
The compounding risk: every task handed to a wizard is a task you did not do, so you slowly lose the expertise you would need to judge whether the wizard's next answer is any good.
Why it matters
It gives practitioners a sharp mental model for working with frontier models like GPT-5 Pro and Claude Opus — recognizing when you're 'summoning a wizard' whose output you can only verify rather than shape, which changes how you design review and verification workflows.
Key quotes
“We're moving from partners to audience, from collaboration to conjuring.”
“We're shifting from being collaborators who shape the process to being supplicants who receive the output.”
“And that suggests another risk we don't talk about enough: every time we hand work to a wizard, we lose a chance to develop our own expertise, to build the very judgment we need to evaluate the wizard's work.”
“The paradox of working with AI wizards is that competence and opacity rise together. We need these tools most for the tasks where we're least able to verify them.”
