How the 10% Work

Last week I made the case that AI's productivity paradox isn't about the tools. It's about how people use them. The 10% who save 20+ hours a week aren't using better AI. They've built different habits around the same subscriptions everyone else is paying for.

This week: the specific techniques I use. If Part 1 was the mindset, these are the habits that make it stick.

1. Start Rough, Get Structured Later

Think out loud, even when your thoughts are not yet well formed. Here's what that looks like in practice.

Don’t hold back. Raw notes from a meeting, a half-constructed presentation, conflicting thoughts about a strategy. Paste it all in (or speak it) and ask AI to help you make sense of it. You know more than you think. AI's job here isn't to generate ideas for you. It's to play your ideas back in a way that reveals what's actually there, including the gaps you didn't consciously catch.

This is especially powerful when you're stuck. Instead of staring at a blank page, talk through the problem out loud (more on that below) or paste in whatever messy fragments you have. Let AI create the first structure. Then react to it. Reacting is always faster than creating from scratch.

2. Brief AI Like You'd Brief a Colleague

If you hired someone brilliant but gave them no context about your work, your stakeholders, or what good looks like, you'd get generic output.

That's exactly what most people do with AI. Last week's first mindset shift was "start with vision, not task." This is how you actually do that.

Before any meaningful conversation, spend 30 seconds loading context (this assumes you haven’t already loaded this information via memory, instructions or files at a project level). Who you are, what you're working on, what the constraints are, what you've already tried. If you read my issue on why context beats prompting every time, this is the AI Context Stack in practice. Your identity files, your state of play, your working materials. They're the difference between AI producing something you'd actually send and something you spend 20 minutes rewriting.

The habit is simple: never start a high-stakes AI conversation cold. Paste in your context first. Every time.

3. Ask for Options, Not Answers

Most people ask AI for one answer. One email draft. One strategy recommendation. One structure. Then they iterate on that single output, tweaking it back and forth until it's acceptable.

Flip it. Ask for five email drafts with different tones. Three strategic approaches with trade-offs. Ten headline options. This feels wasteful, but it's actually faster because you're choosing and combining rather than incrementally improving a single starting point.

This is the tactical version of "Let AI draft, then curate" from last week. The insight is that your taste is faster than your creativity. You'll spot the right approach instantly when you see it alongside four wrong ones. You'd never spot it if you were trying to construct it from scratch.

4. Treat Every Session Like a Shift Handoff

Here's a habit that compounds over time: before you close a conversation, or when you notice quality starting to degrade, ask AI to write a handoff document.

A good handoff captures: the key themes you explored, decisions you made and why, approaches you considered and rejected, open questions, and next steps. Save it somewhere accessible, whether that's a Claude Project, a ChatGPT project, or just a folder on your laptop.

Why rejected approaches matter: when you start a new session and paste in the handoff, AI won't waste your time suggesting things you've already ruled out. You're not starting from zero. You're picking up where you left off, with full context.

And build one more habit alongside this: zoom out regularly. The deeper into the weeds you go with AI, the more valuable it is to pause and ask "Are we still solving the right problem?" Treat it like a project check-in. Reground yourself in what you're actually building before going further.

5. Talk, Don't Type

Voice input is the single biggest speed multiplier most people aren't using.

If "think out loud" was the mindset, this is the tool that makes it effortless. Wisprflow, or the built-in voice features in ChatGPT and Claude, let you speak naturally and get clean transcriptions.

I know it feels awkward at first. Talking to your laptop like it's a colleague takes some getting used to. You'll get comfortable fast, and once you do, you won't go back.

I move roughly 3x faster talking than typing. Not because the words come out faster (though they do), but because talking removes the editing instinct. When you type, you self-censor. You restructure mid-sentence. When you talk, you get your actual thinking out, unfiltered, which is exactly what AI works best with.

Reserve typing for moments that need precision: code, specific numbers, exact phrasing. For everything else, just talk.

6. Stress-Test Your Work Before Anyone Else Sees It

Paste in your strategy doc, your email to the exec team, your proposal. Then ask AI to find the weakest argument, the assumption you haven't justified, the question the CFO will ask that you haven't answered.

Most people use AI to create outputs. Using it to pressure-test those outputs before they reach a human audience is higher value. Ask "What's the strongest counter-argument to this?" or "Where would a sceptic push back?" or simply "What am I missing?"

This is the tactical version of "Challenge AI" from last week, flipped: you're asking AI to challenge you. The result is work that's already survived its first round of scrutiny before it leaves your desk.

7. Use AI to Write Prompts for Other AIs

If you're jumping between tools, and most of us are (Claude for writing, Cursor or Lovable for code, ChatGPT for voice), have your primary AI write the specs and prompts for the others.

Say you've spent 30 minutes refining a product concept in Claude. Now you want to build a prototype in Lovable. Instead of re-explaining everything from scratch, ask Claude to write the implementation brief for Lovable, including the technical requirements, the design direction, and the constraints you've discussed.

It's faster. It's more precise. But always review what it wrote before pasting it across. AI will quietly change scope or make assumptions you didn't agree to. Trust but verify.

Your Checklist

You don't need to adopt all seven habits at once. Start with two:

This week: Try voice input for your next AI conversation. Just talk through a problem instead of typing it. Notice the difference in speed and in how much thinking you actually get out.

Next week: At the end of your best AI working session, ask it to write a handoff document. Save it. Use it to start your next session on the same topic.

These are small changes. But they're the difference between using AI as a search engine and using it as the thinking partner it already is.

The Atomic Builder is written by Faisal Shariff and powered by Atomic Theory Consulting Ltd — helping organisations put AI transformation into practice.