Give AI the First Shot (And Stop Doom‑Scrolling “This Changes Everything” Threads)

Every day, my feeds scream that some new AI model “changes everything.” For a while, I tried to keep up—saving demos, reading every hot take, nodding along like “yes yes, paradigm shift, very cool”—and still feeling weirdly behind. Eventually, I realized the obvious: obsessing over every new feature is an amazing way to feel busy and learn nothing. Using AI on your own boring, real-life tasks? That’s where the actual shift happens.

So, this is my pitch: stop treating AI like breaking news, and start treating it like the slightly awkward new intern you keep forgetting to invite to meetings.

The AI Hype Treadmill (and Why It Feels Awful)

Here’s the cycle I kept seeing:

1. New model drops. Everyone posts screenshots.

2. People declare, again, that “knowledge work is over.”

3. A week later, nothing inside most companies has materially changed.

   
   

Inside businesses, the gap is stark:

• Leadership wants an “AI strategy,” ideally with some futuristic slide in the board deck.

• Teams want fewer meetings, better tools, and less soul-crushing busywork.

• Somewhere in the middle, someone proposes “an AI task force,” which usually means “more meetings, now with buzzwords.”

  
  

The truth is boring and therefore powerful: your AI advantage will not come from being the first one to post a screenshot. It will come from quietly making a handful of workflows 20–50% better, and then doing that again and again.

My One Rule: Before You Work, Let AI Whiff First

If I had to hand you one rule that’s actually changed how I work, it’s this:

Before you do a task the way you always do it, let AI take the first shot.

Not because AI is magical, but because a first pass is often good enough to:

• Break the blank page

• Suggest angles you wouldn’t have considered

• Turn a 90-minute slog into a 25-minute edit session

  
  

Examples from My Day That Are Aggressively Unglamorous

• Prepping for a Client Meeting: Have AI turn a website, LinkedIn, and my notes into a one-page “what this company cares about and where I can help” brief.

• Dreaded Email: Dump messy bullet points and have AI write three flavors—direct, diplomatic, “please don’t hate me”—then mash them into something human.

• New Idea: Ask AI to attack it like a customer, a CFO, and a very tired engineer. It finds cracks I was happily ignoring.

  
  

Is it perfect? No. Is it cheaper than another 45 minutes of my brain chewing drywall? Absolutely.

Beginner Mode: Onboard Your AI Intern

Treat AI like you just hired an intern. If you don’t give them context, you can’t be mad when they act confused.

Step 1: Tell It Who You Are and What You’re Doing

I keep a simple “this is me” block I reuse:

• What I do

• Who my clients are

• How formal I sound when I’m pretending to be professional

• What “good” looks like in my work (concise, specific, not allergic to bullet points)

  
  

Dropping this into a new chat makes the output less generic instantly.

Step 2: Ask AI to Fix Your Own Prompt

Take something you already use a lot—like “summarize this meeting”—and literally ask AI:

• What’s missing?

• What should I tell you up front every time?

• What questions should you ask me before you answer?

  
  

You end up with a reusable prompt that doesn’t suck, built by the thing you’re prompting. Mildly meta. Weirdly effective.

Step 3: Build a Tiny “Context Packet”

For the stuff you do constantly (writing client recaps, internal memos, sales follow-ups), assemble one mini-doc with:

• Goal

• Audience

• Examples of “good”

• Common constraints

  
  

Then, before AI starts, ask it to restate the assignment in its own words. If that restatement is off, you just dodged a bad output.

Intermediate Mode: AI as Your Mildly Annoying Sparring Partner

Once your AI intern knows the basics, it can graduate to “junior employee you side-eye but rely on.”

“Research This Company, But for How I Actually Work”

Instead of “tell me about Company X,” I’ll say something like:

• Here’s their website and a few links

• Here’s my role and why I’m talking to them

• Here are the questions I actually care about

  
  

Then I ask AI for a customized brief: who they serve, likely pains, where my services slot in, what landmines I should avoid. The goal isn’t a perfect dossier; it’s walking into the meeting with more than “so what do you guys do, exactly?”

Ask AI to Be the Critic You Don’t Have

Before I share a new idea, I’ll literally tell AI:

• “Pretend to be a skeptical customer, a CFO, and a competitor. Roast this.”

  
  

It happily obliges, surfacing pricing questions, trust issues, and “this sounds like buzzword soup; what problem does it solve?” style feedback. Is some of it overdramatic? Sure. But it’s cheaper than learning all of that live in front of a room of humans.

Let AI Mark Up Your Website Like a Grumpy Copy Editor

Give it a page, the person I want to persuade, and the action I want them to take. Then ask:

• What’s confusing?

• Where are the claims fuzzy or unproven?

• Which lines are trying too hard?

  
  

You’ll get a brutally honest review and suggested rewrites. You don’t have to accept all of them, but at least you see where a stranger might bounce.

Advanced Mode: Design Systems So Future-You Doesn’t Suffer

If you stick with this, you hit a point where you’re not asking “what can AI do?” but “what system do I want around this work?”

Run Mini Bake-Offs Instead of Arguing About “Best Model”

Take one actual task you care about and do a small experiment:

• Same inputs, same instructions, two or three different models or tools.

• Then ask AI itself to evaluate the outputs against your criteria and point out hallucinations and weak reasoning.

  
  

You’ll quickly learn which tool is best for your use case—not whatever got the most retweets last week.

Turn Every Win into a Reusable Workflow

When something works, I try (imperfectly) to:

• Save the prompt, context, and a “golden” example

• Write two sentences: “use this when…” and “don’t use this when…”

• Drop it into a shared doc or internal wiki

  
  

This is the difference between “oh yeah I did something cool once” and “our team has 10 reliable AI plays we run every week.”

Use a Simple Portfolio: 60/30/10

The way I think about my “AI time” now looks roughly like:

• 60% running the plays that already work

• 30% improving those plays—better context, better prompts, better checks

• 10% trying weird new things that might flop

  
  

That last 10% protects you from getting stale, but the other 90% is where the compound gains live.

The “Not Yet, But Soon” List

Some tasks will still be bad fits—today. Maybe the model keeps hallucinating, or the data you’d need is locked in eight incompatible systems and one rogue spreadsheet. Instead of declaring “AI can’t do X,” I keep a small “not yet, but soon” list:

• Tasks that almost worked

• Why they failed (data gap, reasoning, tooling)

• A note to retry in a few months

  
  

The models keep getting better whether we’re paying attention or not. The point is to have a short list ready for Future You to revisit without starting from scratch.

Start Small, Start Real, Start Today

If you want to feel “caught up” on AI, here’s a radical idea: ignore the next three “this changes everything” posts. Pick one task you already plan to do today, and let AI take the first swing. Then:

• Keep what helped

• Fix what didn’t

• Save it if it worked

• Try again tomorrow with something slightly harder

  
  

That’s it. That’s the game.

What’s one very real, very unsexy task on your plate this week that you’re willing to let AI whiff on first, just to see what happens?

---wix---