Should you use AI for a task? A Harvard professor has a simple test

Key points

• Bruce Schneier teaches public policy at Harvard Kennedy School and the University of Toronto's Munk School.
• His students regularly use AI to complete writing assignments, which he argues wastes their tuition investment.
• AI researcher Daniel Meissler's "work versus gym" framework offers a clean way to decide when AI help hurts you.
• The key question is not whether AI can do the task, but whether doing it yourself is the point.

Bruce Schneier has a problem shared by professors everywhere. His students, enrolled in some of the most competitive public policy programmes in the world, hand in AI-written essays.

He teaches at Harvard Kennedy School and at the Munk School of Global Affairs and Public Policy at the University of Toronto. They're paying considerable fees to learn. Then they quietly hand the thinking to a chatbot.

Why does that matter if AI is their future?

The writing was never really the product. The thinking was.

Schneier, writing in The Guardian, borrows a framework from AI researcher Daniel Meissler to explain the trap. Meissler draws a line between work and the gym.

At work, output is what counts. If software can process your invoices faster, use it. The invoice doesn't care how it got filed.

The gym is different. You don't hire someone to lift weights on your behalf and expect to get stronger. The struggle is the mechanism, and skipping it means skipping the benefit.

Writing, both men argue, is often a gym activity dressed up as work. We covered a related tension on 18 July when Australian universities split over whether AI-assisted assessment was cheating or preparation, with one expert warning that weak standards could cost the country its intellectual talent.

So when should you actually use AI?

Before you start, ask: is the finished product the point, or is producing it the point?

Formatting a report or summarising a meeting you already attended puts the output first. AI is a reasonable tool for that.

Trying to learn to argue clearly, or building the kind of judgment that gets you promoted in a decade, puts the process first. Handing that process to a large language model (the technology behind chatbots like ChatGPT) means handing away the benefit you came for.

The same logic applies outside the classroom. A nurse writing incident reports for compliance is doing work. A nurse using AI to avoid learning how to document clinical observations is skipping the gym. That distinction is worth keeping in mind whenever a task feels like a chore: the chore might be the point.

What should students and professionals actually do?

Schneier doesn't say never use AI. He says be honest about what you're giving up when you do.

For students, that means treating AI as a checking tool or a starting prompt, not a ghostwriter. Draft the argument yourself, then ask the AI where it's weak. That order matters.

For anyone building a skill, the same discipline applies. Fluency comes from repetition, and there's no shortcut that leaves the fluency intact.

The framework is almost insultingly simple once you hear it. Work versus gym. Output versus process. It won't stop students submitting AI essays, but it gives anyone willing to think honestly a clear place to start.

Common questions

Does this mean AI tools are bad for learning?

Not automatically. AI can be a useful checking or feedback tool when you do the underlying thinking first. The problem is using it to skip the thinking entirely.

How do I know if a task is "work" or "gym"?

Ask whether completing it yourself would make you better at something you care about. If yes, do it yourself. When the only thing that matters is the finished result, AI assistance is reasonable.