Start Simple, Stupid

Pretty much every AI success story I've been part of started embarrassingly simple.

The vibe was always less "transform the company", and more like "get rid of the annoying things". Like a simple chatbot that answers the same five questions that get asked over and over in unlimited variations. Or a workflow that pulls data from one system, applies some little intelligence, and dumps it into another. Things that are so boring nobody would even put them on a slide.

Starting simple and boring is probably the best AI advice there is.

But there's a second part that most businesses get wrong. They start simple and boring. But they stay isolated. The simple thing becomes its own little island. And six months later, you're not building anymore – you're managing little islands.

Today's article is about how to start simple without building a junk drawer.

Let’s dive in!

The Junk Drawer

You know that drawer in your kitchen? The one with the old charger, the lighter, a pack of disposable chop sticks and three pens where only one is (more or less) still working. Everything in there was useful at some point. But nobody really remembers when it stopped being useful. And of course, nobody wants to clean it out.

That's what AI looks like in a lot of organizations after 12 months.

A chatbot here, an n8n workflow there, a Slack summarizer someone vibe coded during a hackathon. Or maybe just a document extraction tool someone got from a vendor that one team now swears by and nobody else knows exists. Multiple tools, built in isolation, none of them connected to each other. The way that led here was always the same: Someone had a problem, found an AI solution, shipped it, moved on. Repeat ten times.

Now you have ten islands and no bridges.

The frustrating part is that none of these were bad decisions individually. Starting small and low-risk is smart. The problem isn't the starting point – it's that there was never a second step in mind. The sequence was missing.

I've written before about why First AI beats AI First. But even companies that get that part right often still end up with a junk drawer. Simply because the individual projects weren’t designed to connect.

Starting simple is right. But starting simple and isolated is how you end up with a junk drawer that’s waiting to get cleaned up.

Dead Ends vs. First Floors

So if simple is the right starting point, what separates the junk drawer from actual progress?

In short, one question:

Did the simple thing make the next simple thing possible?

A dead end solves one problem and stops, because "it works". Maybe it works even so well your organization will eventually forget about it. You’ve just created another island. Because this island didn’t lead anywhere. No data you could use later on. No infrastructure that other tools could plug into. No organizational capability that would unlock the next "bigger", and suddenly simple thing.

A first floor starts at the same point by solving the same job. But on top of that it creates a condition for what comes after. It generates or unlocks new data, reveals patterns, and informs your next step by allowing you to transfer the learnings you’ve made or the capabilities you’ve built to something else.

The difference is in the intent.

To make this concrete, let’s zoom in on that simple FAQ chatbot from the intro. You can build it two ways:

Version A: It can answer the most common customer questions.

Version B: It can answer the most common customer questions, but it also allows you to tap into the chatbot data and track what people have asked, how they phrased it, and where it can't help. It also allows you to swap the chatbot interface for any touchpoint you like (email, messages, perhaps even phone calls).

It’s the same user experience on day one with similar efforts to build. But version B built a first floor, and version A built a dead end. Six months later, version B will know exactly which steps to automate next because the data is already there. Version A will still be just answering FAQs.

Before You Build

To be super clear: this isn't just about chatbots.

Last week I wrote about why AI roadmaps beat AI projects — why mapping your total AI opportunity before building anything changes the entire trajectory of your AI initiative. The core idea: when you have a map, no single project has to justify the whole investment. Each one is a step, not a standalone bet.

An AI roadmap only works if the projects on it connect. If every item on your map is its own dead end – its own isolated quick win island – you don't have a roadmap, just still a junk drawer. (A slightly more organized junk drawer, perhaps.)

The "first floor" question is what turns a roadmap into a profitable AI system.

• A chatbot that answers the most common questions but also tracks topics and user intent so you can build systems that eventually make customers never raise a support ticket at all.

• A document extraction tool that logs which document types come in and which fields cause errors so you can implement auto-validation rules or sub-workflows for different document types altogether.

• An email summarizer that assigns topic and tags which paves the way for automated routing and drafting answers.

Each project feeds the next, and each first floor makes the second floor cheaper, faster, and easier to reach (and therefore easier to justify).

That's the difference between an AI roadmap where every project fights for its life, and an AI roadmap with compounding momentum.

Conclusion

So here's your action item for the next AI project on your list — whether it's a chatbot, an extraction tool, or something entirely different.

Before you build it, ask: "Does this make the next thing easier, or is this where it ends?"

If the answer is "this is it" — you're building a toy. It might work. It might even save time. But six months from now, it'll sit in the junk drawer with everything else.

If the answer is "this feeds into X" — even if X is vague right now, even if you're not sure what the second floor looks like yet — you're building a system. And that one question will change what you build, how you build it, and what becomes possible after.

Start simple and stupid. But be smart about it.

See you next Saturday,
Tobias