AI First vs. First AI

It happened again when I was recovering from the Christmas feast on the couch. Scrolling a bit through my feed, I spotted another company proclaiming to become "AI First".

The big transformation. I was curious, so I checked their website.

I found a contact form that looked like straight out of 2001. Zero signs of applied AI. Not even a simple chatbot for Q&A.

"AI First".

While I think that becoming AI First can be a great goal for a lot of companies, I've realized that for most established businesses it’s actually a terrible ambition level to start.

Why I think that’s the case and what alternatives there are is part of today’s article.

Let’s dive in!

The Two Patterns

After working with dozens of companies on AI implementations, I've noticed two distinct patterns.

Business A has a great, ambitious AI plan and plenty of "roadmaps". They spent a lot of time thinking about their "AI Vision" and hired a bunch of external consultants to create a 50+ strategy deck. There are plenty feedback loops, before the first thing gets built. The pattern looks like a spiral:

Plan → Plan → Plan → Refine → Refine, followed by some occasional, longer "Build" bursts which are then wrapped again in longer planning cycles. God knows what’s next.

I call this the AI First pattern. The wheel keeps spinning, but it never crosses the ambition line.

Even after a year in, you feel stuck because reaching your ambition level feels still so far away.

Then there's another business.

Business B.

Business B sat down to identify their high-level goal as well. It's what I call their AI adoption approach, figuring out whether you're in for the R&D or the ROI. From there, they identified their minimal ambition level (see 10K Threshold) to actually get started.

Then, it's a series of smaller Plan → Build → Learn → Refine loops which looks like this:

Each loop crosses that minimal threshold. Then you do it again, slightly bigger. The loops compound. You're not trying to transform everything at once, but trying to cross a finish line. Then the next.

I call this the First AI approach.

The goal being to ship your first Engineered AI solution that actually matters and has the chance to compound into something bigger.

Avoiding Transformation Theater

I've seen this movie before. It was called "data-driven". Before that, "digital-first". Great ideas, but terrible execution.

Companies wanted the transformation so badly that they skipped the part where you actually build things. I used to work for an organization that had "we want to become a data company" on their board slides while they couldn't ship a simple dashboard in under 6 months. That was peak "data" hype cycle.

Now it’s happening again with AI.

Transformation theater where you spend more time on ideas and concepts than actually building and maintaining things.

"We Need to Get Ready First"

Here's what companies tell themselves:

We're not ready yet.
We need to get the right ownership in place.
The right technical knowledge.
The right tech stack.
Cultural buy-in.
A proper data foundation.

Fair enough. All of these things matter.

But here's the problem: waiting doesn't make any of them easier.

You won't magically develop AI expertise by planning. You won't get cultural buy-in by writing strategy docs. You won't know if your data foundation is good enough until you try to build something on it.

AI has too many moving variables to get everything right upfront. The goal isn't to avoid all mistakes. It's to avoid irreversible mistakes. There's a big difference.

You have to make a plan before you start. "Planning replaces chance with error", as my friend Jürgen rightly pointed out. But plans are the means, not the goal. At some point, you have to cross the threshold. Transformation theatre won't take you there.

So it’s all about Quick Wins?

Now, some companies try to escape the planning spiral with "quick wins":

• Marketing buys an expensive "Social Media Content for Insurance" SaaS.

• Legal gets a "Value Extraction from Contracts" tool.

• Someone signs up for yet another AI productivity suite.

This looks like progress. It isn't.

These isolated purchases create vendor lock-in and make you exactly 0% more AI-ready. They're not loops that compound. They're bandaids.

Quick wins only matter if they can compound into something bigger. Otherwise, they're just expensive distractions — unless your AI adoption approach is purely opportunistic.

It’s all about First AI

So what's the alternative?

First AI means building one relevant Engineered AI solution that works profitably. Not a prototype or a showcase. Something that lives in production and delivers real value every single day.

Because when you do that right, you're not just solving one problem. You're building an AI asset – something that gives you more options and grows as you go.

Let me show you what First AI looks like in practice.

Example 1: The Call Center

A client (small call center company) wanted to explore the idea of replacing human agents with AI support voicebots. That's a big ambition — and a recipe for a two-year roadmap with relatively few chances of success (if you try to do it all at once).

Instead, we started building a simple AI-search tool that helped human call center agents find answers faster. That was loop one. Shipped it. Learned from it.

Loop two: an AI system could listen to calls live. The transcripts would be used for post-call evaluations.

Loop three: Now that the AI would listen live and we had a pretty good search function, we could build an augmentation that suggests answers to agents in real-time.

Loop four: it starts giving answers directly (online chatbot, text instead of voice).

Each loop crossed a new threshold. Each one built on the last.

Today, they still haven’t got that voice chatbot yet — but they’re happy with the compounding wins that got them here.

Which also made them realize that they prefer having humans on the phone, but making the whole onboarding and training experience much cheaper and faster thanks to their AI assets.

Example 2: The AI Mini-Me

A consultant wanted to productize his expertise with an "AI mini-me" — a tool that could handle client work the way he would.

Big vision. But instead of planning it all out, he started with one small end-to-end task: creating offers. One loop. Shipped it. Validated that clients actually wanted this. Validated that he could vibe-code the app and maintain it himself.

Now he's expanding to the next loop. The only roadmap he needed was the confidence that building the “little skills” could later be integrated or compounded in one “main system” – if he ever wanted to.

The difference to a “quick win” is that these aren't just tools. They're AI assets that grow as the business evolves.

Conclusion

I don't want to make "AI First" sound wrong. It's not.

AI First can be a great destination. If you genuinely want your entire organization to be AI-native, that's a valid ambition. Some companies need comprehensive roadmaps because the opportunity size justifies it.

But what I've learned is that going directly for AI First is often counterproductive – your ambition level crushes your execution muscle.

Roadmaps are an integral part of your AI journey. But they’re not the goal. The goal is shipping something real that has the potential to transform your organization as you go.

Sometimes, you don't need to see the whole staircase to take the first step. You just need to see the step.

See you next Saturday!
Tobias