The 30-Day AI Experiment You Can Run Without a Budget or a Blessing

Theme 3, Article 8 — 2weekAI Blog

At some point in the last seven articles you've probably had a version of this thought.

"This all makes sense. But I don't know where to start. And I don't have time to figure it out. And nobody has approved anything. And my IT department—"

Stop.

You don't need a budget. You don't need approval. You don't need a transformation initiative or an AI committee or a working group that meets monthly and reports back in Q3.

You need thirty days, one problem, and enough curiosity to see what happens.

That's it. Let's build the experiment.

The Setup: One Problem, One Tool, One Month

The mistake most people make when they try to start with AI is trying to start with AI.

They download five tools, watch three YouTube tutorials, read a LinkedIn post about prompt engineering, get overwhelmed, and go back to the spreadsheet.

Don't do that.

Pick one problem. The most annoying recurring task in your week. The one you identified back in Article 4. The status report. The data pull. The meeting recap. The thing you've done a hundred times and could do in your sleep.

One problem. That's the whole experiment.

Week 1: Understand It Before You Automate It

Before you touch an AI tool, do something most people skip.

Write down exactly what the task requires.

Not the output. The inputs. What information does this task need to get done? Where does that information come from? Who touches it before it gets to you? What decisions do you make while doing it that aren't obvious to someone watching from the outside?

This is the Article 5 question applied in practice. What is this task actually for? What question is it trying to answer? Is that still the right question?

Spend a day on this. Just observation and documentation. No tools yet.

By the end of Week 1 you should be able to describe your task to a stranger in five minutes and have them understand exactly what done looks like.

If you can describe it that clearly to a stranger, you can describe it that clearly to an AI.

Week 2: Run It in Parallel

Now introduce the tool.

For most recurring tasks involving writing, summarizing, or organizing information — start with Claude or ChatGPT. Free tier is fine for this experiment. You're not deploying a system. You're testing a hypothesis.

Run your task the normal way. Then run it again with AI assistance.

Don't try to hand everything off on day one. Treat it like a new team member's first week. You're not delegating yet. You're training.

Give the AI the context it needs. The format you use. The audience you're writing for. The three things that always have to be true about the output. The one thing that always gets wrong if you don't specify it.

This is where your institutional knowledge earns its value. The AI is capable. You're the one who knows what capable looks like in this specific context.

Run it in parallel for the full week. Compare outputs. Note what it gets right immediately. Note what needs correction. Note what it misses that you had to add.

That gap between what it produced and what you actually needed — that gap is your prompt. Write it down.

Week 3: Refine the Handoff

By Week 3 you have something valuable.

You have a description of the task. You have the context the AI needs. You have a list of what it gets right and what it misses. You have the corrections you made and why.

That's a prompt. A real one. Built from actual use, not theory.

Spend Week 3 tightening it.

Every time the output needs correction ask yourself: could I have prevented this with better instructions upfront? Usually yes. Add it to the prompt.

By the end of Week 3 you should be getting outputs that need minor refinement instead of significant rework.

The task isn't automated yet. But the heavy lifting has moved.

Week 4: Measure the Actual Difference

This is the week most people skip and it's the most important one.

Document what changed.

How long did the task take before? How long does it take now? What did you do with the time you got back? What's the quality of the output — better, same, or needs more refinement? What would it take to get this to the point where the output is ready with minimal review?

Write this down. All of it. Not for anyone else yet. For you.

Because this document becomes two things.

First, it becomes your personal proof of concept. The next time someone says AI can't do real work, you have thirty days of evidence that it can — in your specific context, on your specific task, with measurable results.

Second, it becomes the business case for the next conversation. Which we'll cover in Article 9.

What You're Actually Building

The experiment isn't really about the task.

The task is just the vehicle.

What you're actually building in thirty days is a new thinking pattern.

The habit of asking what the task is for before you do it. The discipline of documenting context so AI can use it. The instinct for where AI adds value immediately and where it needs more direction. The confidence that comes from having done it once and seen what happened.

That's the thing that compounds.

The first task takes thirty days. The second takes two weeks. The third takes three days. By the time you're six months in, you're assessing new tasks for AI potential before you've done them once.

That's the thinking pattern shift that Article 5 was pointing at. This is what it looks like in practice.

The Lowest Stakes Experiment You'll Ever Run

Here's the honest risk assessment.

Worst case: thirty days from now the AI outputs aren't good enough, you've learned something about your own task that you didn't know before, and you're back where you started.

You've been there every week for years. That's a very manageable downside.

Best case: thirty days from now you've reclaimed hours every week, you have documented proof that AI works in your specific context, and you have a repeatable process that keeps improving.

And somewhere between those two outcomes is the thing that actually happens to most people who run this experiment.

They finish thirty days and immediately start looking at what else in their week has AI's name on it.

Because once you've seen it work once — really work, on a real task, in your real job — you can't unsee it.

That's the experiment. Thirty days. One task. One tool.

The worst outcome is you learned something. The best outcome is you changed something.

Both are worth the thirty days.

A Note on the Craft

The best leaders I've worked with over 28 years never lost the craft.

The VP who could still take the call. The engineer who could still write the code. The executive who could still sit in the seat and do the work.

Title added responsibility. It didn't replace capability.

The leaders who figure out AI aren't going to be the ones who delegated it to a committee. They're going to be the ones who ran the experiment. Who got their hands in it. Who understand from direct experience what it can and can't do.

Because you can't lead what you don't understand.

And you can't understand it from a briefing deck.

Thirty days. One task. Go find out for yourself.

*Last one: How to bring this to leadership without getting laughed out of the room. The business case, the framing, the timing, and how to use your thirty days of evidence to start a conversation that actually lands.*

*2weekAI deploys AI that actually works — in 2 weeks, at a fixed price. Dave has delivered AI solutions across Fortune 100 enterprises and growing businesses alike. No transformation program required. [Book a discovery call.]*