The question most content teams are still asking about AI is the wrong one. Whether AI can write as well as a human misses the point entirely, and the framing keeps organisations stuck in a philosophical debate while a practical productivity problem goes unsolved.
Christopher Nolan said as much in a Telegraph interview, calling the ‘AI replaces human creativity’ narrative «nonsense.» His own use of AI runs to technical fixes: removing stunt wires in post-production, correcting lighting in shots. He does not ask it to construct a story. That distinction, from someone whose craft depends on originality, is more useful to content teams than any vendor briefing.
The reason the wrong question dominates is partly structural. AI tool vendors have a commercial incentive to promise that their product can do the hard creative work. Coverage follows the drama of that claim. Meanwhile, Harvard Business Review contributor Andy Wu built a quieter, more functional framework in his Gen-AI Playbook that most marketing teams have not acted on: stop asking whether AI is as creative as a human, and start asking whether it can save time and make creative work easier. Those are different questions with different answers.
The framework that actually sorts the problem
Wu’s framework splits creative tasks into two categories. Explicit-knowledge tasks involve applying rules, formats, and retrievable information: AI handles these well. Tacit-knowledge tasks require empathy, lived experience, and professional intuition built over years: these stay human. The division sounds clean in theory. In practice, content teams blur it constantly because no one has sat down and sorted their actual task list against it.
Nolan offers an unintentional illustration of what tacit knowledge looks like at the top of a profession. He describes writing as «very hard, and very lonely,» and notes that even he needs external triggers to get unstuck: walks, music. The blank page problem is not solved by intelligence alone, human or machine. It requires something closer to psychological navigation. AI cannot replicate that, but it can, as Wu suggests, act as a creative catalyst: generating multiple opening paragraphs for a specific audience, brainstorming title variants, proposing analogies. These are explicit-knowledge assists for a tacit-knowledge process.
The strongest argument against drawing a firm line is that tacit knowledge is eroding faster than expected. Models trained on vast bodies of professional writing are already producing copy that passes casual inspection. Some marketing directors will read Wu’s framework and reasonably ask whether ‘empathy and intuition’ are truly irreplicable or whether they are simply harder to automate on a short timeline. That scepticism is worth holding. The answer is not that AI will never close the gap, but that right now, in 2025, the gap matters on client work where brand voice, audience trust, and relationship context are load-bearing.
What this means for task allocation on client accounts
Inc. covered Nolan’s comments and Wu’s framework together, and the combination points toward a concrete exercise most content teams have skipped: a task audit using explicit versus tacit as the sorting criteria.
Explicit-knowledge tasks suitable for AI delegation include formatting content to house style, generating three to five headline options from an approved brief, producing a first-draft outline from a defined structure, transcribing and summarising interviews, and suggesting analogies or examples to illustrate a concept that the writer then tests and refines. These tasks consume time without requiring the contextual judgement that defines professional content work.
Tacit-knowledge tasks that should not be delegated on client accounts include deciding what a brand should say about a sensitive topic, writing the opening paragraph of a piece where tone sets the entire client relationship, editing for voice consistency on accounts where the brand persona has been built over months, interviewing subject-matter experts and extracting the insight that is not in the transcript, and judging whether a finished piece will land with a specific audience at a specific moment. These are not tasks AI handles badly because of a technical limitation. They are tasks where the cost of getting it wrong is client trust, and where no tool has the account context to make the call.
The practical implication is straightforward. Print Wu’s two categories. Take your content team’s standard task list for one client account. Put each task in a column. Any task that lands consistently in the explicit column and still occupies senior time is a workflow inefficiency with a fix available now. Any task that lands in the tacit column and is currently being pushed toward AI tools to save time is a risk that has not been priced into the retainer.
Nolan removes wires. He does not ask a model to decide what the scene means. Content teams that have not made that distinction explicit are leaving both efficiency and quality on the table simultaneously, which is a harder position to defend to a client than either one alone.
