Scene hook
Last night I was revising a client proposal at my computer. The AI answered in 5 seconds, but the more I read it, the less confident I felt. I’ve fallen into this trap too: it wasn’t that the model couldn’t do it—I just hadn’t let it “think a little longer.”
What this method is + who’s already using it
I only really understood this after seeing Sebastian Raschka talk about “controlling reasoning effort.” With the same AI, it’s not just about switching models—you can also adjust how much time it’s willing to spend thinking. In simple terms: email replies and headline tweaks can stay fast; quotes, requirement breakdowns, and client proposals can go slower and think deeper. Last Tuesday at 9 p.m., in a coworking office in Binjiang, Hangzhou, Aran, a brand consultant, was holding a coffee and switched an “event recap” task into a higher-thinking mode on the spot. What had started as vague conclusions immediately turned into three concrete actions someone could actually execute. That moment really won me over, and honestly made me regret how long I’d optimized only for speed.
What it costs to replicate today
Money: RMB 0-20 a day (most tools let us test this on a free plan; only heavy use usually costs money). Time: 15 minutes. Technical barrier: almost no coding—you mostly just need to know how to switch options like “fast” or “deeper thinking” inside a chat tool. First step: open the AI chat tool you already use, click the button next to the model that says something like “think more,” “deep thinking,” or similar, then paste in one real client question and compare the fast answer with the slow one. Not everyone needs this. If most of what we do right now is everyday copywriting, it’s fine not to bother yet.
Advice by stage: just starting / 1-2 clients / scaling up
If I were just starting, I’d use it first for pricing, service descriptions, and self-intros instead of going straight into researching a pile of new tools.
If I already had 1-2 clients, I’d run a real A/B test on actual requests: ask the same question twice, once in fast mode and once in slow mode, then see which version is closer to something I could send to a client directly.
If I were scaling up, I’d write down a small team rule for which questions use fast mode and which use slow mode—for example, customer support scripts need speed, while first-draft proposals need stability. For small teams like ours, the gap often isn’t tools. It’s making this judgment call in the first place.