What AI can and can't do for productivity

AI productivity advice is drowning in hype, so let's start with what the actual evidence shows — because it's more nuanced (and more useful) than 'AI makes everyone 10x productive.' Getting an honest picture is what lets you capture the real gains and avoid wasting time.

The gains are real — on some tasks. Controlled studies show genuine productivity gains for AI on well-defined knowledge tasks: business writing has been measured at roughly 40% faster with higher quality; customer-support workers gained meaningfully; consultants working within AI's reliable zone completed more tasks, faster, at higher quality. So for drafting, summarizing, and well-scoped tasks, the time savings are real and documented.

But the gains vary enormously — and can reverse. Here's what the hype skips. The same research shows the benefit is highly uneven:

• Novices gain most; experts gain little. In support work, the least-experienced gained a lot while the most-experienced saw roughly zero. AI often levels up the less-skilled more than the already-expert.
• **On the wrong tasks, AI *slows you down.*** In one study, experienced developers took ~19% longer with AI tools — while believing they'd been sped up.
• At the macro level, gains can vanish. A large study across many occupations found AI chatbots had no significant impact on hours or earnings on average — people felt meaningful time-saved, but on average it didn't translate into fewer hours worked or higher earnings (in part because saved time gets absorbed by new tasks, and by checking and fixing AI output).

The perception trap — the most important thing to know. Across studies, a striking pattern: people consistently feel more productive with AI than they actually are. Developers felt faster while being slower; workers overestimated their time savings. Your gut sense of 'AI is saving me tons of time' is unreliable. This is why the course emphasizes measuring real results, not trusting the feeling.

What this means for you. AI genuinely helps — but not automatically, and not everywhere. It helps most on well-defined tasks, especially drafting and summarizing, and especially if you're not already expert at the task. It helps least (or hurts) on expert work, high-stakes judgment, and tasks outside what it reliably handles — and the time you 'save' can be eaten by verification. So the goal isn't 'use AI for everything'; it's 'use AI where it genuinely helps, verify what matters, and measure whether it's actually saving you time.'

The takeaway: the honest evidence on AI productivity is nuanced. Real, documented gains exist on well-defined tasks — business writing ~40% faster, support and consulting work improved — especially for drafting/summarizing and for non-experts. But the gains vary enormously: novices gain most while experts gain little, on the wrong tasks AI actually slows you down, and macro studies find average gains can vanish (eaten by verification). Most importantly, people consistently feel more productive than they are — your gut is unreliable. So use AI where it genuinely helps (well-scoped drafting and summarizing), verify what matters, and measure real results instead of trusting the feeling.