Being an effective AI whisperer

AI tools are rapidly becoming our collaborators. And one thing is clear: the way we communicate with these tools determines the quality of what we get back. This matters more than you might think, if you want to leverage AI effectively.

Soft skills are becoming hard skills

Clear communication is perhaps the most important soft skill of all. But with language-based AI now baked into our work, its importance has increased even more.

Why? Because articulating our thoughts with precision is no longer only about getting a human colleague on the same page. It's about guiding a powerful, yet often too literal, digital mind.

Think about it: when I'm explaining a task to a long-time teammate, we have a ton of shared context: our team's coding practices, project history, the unspoken "how we do things around here." AI doesn't always have all that. I can't assume it knows about that specific library we always use or the nuances of our design system. I have to be more explicit, more structured.

And then there's feedback. Providing feedback to AI is different than with a human. For one thing, we can be more direct. "Change X to Y" is fine. But being skilled here isn't only about being direct; it's about being direct in a way the AI understands and can act on effectively. It's less about emotional intelligence and more about instructional precision.

Strategies for brainstorming with AI

Here's what I've found useful when I start working with an AI with a complex goal in mind.

The strategic pause: "don't jump to solutions yet"

I lay out the problem, give it the context I think it needs, and then I almost always end with something like, "don't jump into solutions yet".

Why the pause? Mainly, I want to see the options, and I want the AI to consider them all too, instead of picking one option and start working on it. Also, the AI might suggest approaches I hadn't considered, or, it might propose something that I know won't fly because of context I haven't explicitly stated yet.

Seeing its initial take on "what could be done" allows me to course-correct early, or even adopt a suggestion I hadn't thought of. It prevents the AI from churning out a fully formed solution down a path that's a non-starter. This initial exploration also helps clarify my own thinking, acting like a smarter rubber duck.

The clarity check: "ask me questions"

Another move I frequently make is to explicitly ask the AI to ask me questions. "What am I not seeing here?" or "What else do you need to know to help me better with this?" This is invaluable.

The AI's questions can highlight ambiguities in my own thinking or reveal aspects I hadn't considered. It forces me to articulate my needs more clearly, and in turn, helps the AI understand the task on a deeper level.

This dialogue is key to getting more refined and relevant outputs. It's a bit like turning the tables. I'm not just prompting it: I'm prompting it to prompt me for better clarity.

The focused restart: "draft our next prompt"

After a good amount of back-and-forth, especially if we've explored multiple avenues or brainstormed various ideas (common when working on code, for instance), the conversation can get a bit messy with details that are no longer relevant.

At this point, I'll often ask the AI something like "Okay, based on everything we've discussed, summarize the specific plan we've landed on into a new, clean prompt for yourself". Here's the exact version I have as a snippet in Raycast:

Summarize the final plan we've agreed on into a clear, standalone prompt. Be specific about the chosen technologies, libraries, and any key details needed to execute it. Ignore any discarded ideas. This prompt should be complete enough to start a fresh conversation without relying on prior context.

This new prompt effectively filters out all the exploratory dead-ends and focuses only on the chosen path. I then take this AI-generated summary and start a fresh conversation with it. This ensures the AI isn't "tainted" by previous, now irrelevant, context and can execute the refined plan with maximum clarity.

Bonus strategy: voice dictation instead of typing

This one's not so much about improving AI output as it is about communicating more efficiently: use voice dictation software to quickly capture your thoughts as text.

It is more efficient if you talk to the AI, but to have the AI answer you with text, and not voice. Dictation software enables just that. Plus, you can use it for other purposes too. This has been a game changer for me.

There are various apps out there that make this very easy. My two favorites are wisprflow.ai (referral link, not sponsored) and superwhisper.com.

Key takeaways for effective AI dialogue

Here's a distilled version of the above strategies:

1. Prioritize exploration over execution (at first)
2. Turn monologue into dialogue
3. Iterate towards precision during the brainstorming phase
4. Start fresh when switching to execution
5. Use voice dictation instead of typing

Try these out, change them to your liking, and make them your own.