Some tasks are worth slowing down for. Getting them right matters more than completing them fast and missing the outcome you wanted.
Communication is one of those tasks, one of the highest-value parts of running a business day to day. It's also one of the hardest to start: a blank page doesn't write itself. This is where AI helps, not just to move faster, but to raise the quality of what you send.
The technique I like most is the rule of three: instead of taking AI's first draft, ask for three options and then use your own judgment.
Core insight: On high-value tasks, don't take AI's first answer. Ask for three options and choose deliberately, the same way you'd think through a decision yourself.
The workflow
Let's walk through how we'd follow this using communication as an example.
Step 1: Get one real fact about the person, and know where it came from. There are two ways to do this, and neither is wrong.
Manual: you already know something, a LinkedIn post, a funding announcement, a line from a meeting. You hand that fact to AI yourself.
Automated: if your AI is connected, meaning it can search the web, or it has access to your Slack, email, and calendar, ask it to find the fact for you. "Find one recent, real fact about [Name] at [Company] I could reference in an opener" works, and a connected AI can also look inside your own history with this person: what you last discussed, what their team currently cares about, what came up the last time you talked. Either way, the fact needs a source attached. If AI surfaces it, it should say where it found it. Check the citation before you use it, since a wrong fact is what actually damages the relationship.
Let's take a look at three different scenarios, depending on what type of business or organization you're in, or what role you play within it:
Cold outreach (a prospect or new contact): LinkedIn activity, a recent funding announcement, a product launch, a news mention, or a data pull from a tool like Apollo. Fully searchable, so this is the easiest one to hand to a web-connected AI end to end.
An investor you're trying to raise from: their public thesis or writing, a portfolio company that overlaps with yours, a podcast appearance, or a mutual connection who can vouch for the intro. Also searchable, though whether a given portfolio company actually overlaps with your thesis is still a judgment call you make, not AI.
An internal stakeholder (a colleague, another team, a cross-functional partner): the last meeting note, a Slack thread, their team's current priorities, or something they said in a standup. This is where a connected AI has a real edge, since it can pull straight from your own Slack, email, and calendar instead of you digging back through history yourself.
Step 2: Run this prompt, swapping in your variables.
Write 3 different one-line email openers to [Name], who is [Role] at [Company/Team]. Reference this fact about them: [the fact]. Make one warm and conversational, one direct about the outcome I want, and one that asks a genuine question. Keep each under 20 words.
Step 3: Read all three like an editor, not a customer. Which one sounds like something you'd actually say out loud to this person? Which one gets the fact right? Which one matches how well you actually know them? Pick that one.
Step 4: Drop it into your existing message and tweak it as you see fit. The rest of the email, the ask, the close, still comes from you.
Here's how that plays out across three illustrative situations:
Notice the skeleton never changes. Warm, direct-on-outcome, question. Only the fact and the stakes change.
The ROI unlock
The honest version of the time saved: you're not saving hours here, you're saving the 10 to 15 minutes of staring at a blank line per message. You're also saving the much larger cost of a bad opener, the one that gets deleted, or the internal one that quietly damages trust because it read like you didn't do your homework. Across a week of outreach, that's the difference between a handful of real replies and a folder of ignored messages.
💡 Key takeaway: On the tasks that matter most, don't take AI's first version. Ask for three, then choose like an editor, not a customer.
Tool worth a look: Mira
Mira, by Decode/Entropik, runs the entire research workflow end to end: recruits participants globally, moderates AI-driven interviews with dynamic follow-up probing, then synthesizes themes and a report automatically. What sets it apart is it also reads emotion in real time through facial coding, voice analysis, and eye tracking, catching the gap between what someone says and what they actually feel.
I haven't tried it myself, but it's a clean example of a pattern I keep seeing in AI-native tools: redesigning the whole workflow instead of automating a single step.
Found this useful? Forward it to one person who needs it. Getting this for the first time? Subscribe free here.
Think Better. Execute Faster.
See you next week - Purti
Clarity Prompts - The weekly AI operating system for sharper leaders.
