The Freeze Test

Today’s a quick one.

I'm still recovering from a packed week — flew out to Paris for Nvidia GTC, came back around midnight, slept maybe five hours, and jumped straight into delivering a 6-hour online training. Power outage in the middle of that? Sure, why not.

But honestly – none of that matters when you’ve just met Taylor Swift, but for tech:

(Still processing – more from GTC Paris next Friday!)

Anyway — today’s email is a quick deep dive on a method I use to sanity-check AI projects. Basically: a simple litmus test to figure out if what you’re building is durable, or just surfing the latest wave.

Let’s talk about the Freeze Test.

Why chasing the frontier might be the wrong strategy

In AI, there’s a near-religious belief that you have to be at the cutting edge at all times.

“What’s the latest model?”
“Did you see that new feature drop?”
“Is your team already on o3-Pro?”

That’s the daily heartbeat of the AI startup scene. And the underlying assumption is clear: To stay competitive, you have to keep up. Every week. No exceptions.

But I believe that’s backwards.

The real test isn’t what’s new – but whether your solution would still be useful if nothing changed tomorrow.

Enter: The Freeze Test

The Freeze Test

Here’s the test every AI project on my desk needs to pass:

Would people still use it? Would someone still pay for it? Would your team still rely on it day to day?

If the answer is yes — great. You’re building something durable. Something grounded in a real, repeatable need. You’re using today’s tools to solve today’s problems — and anything that comes next is just upside.

But if the answer is no, you might be caught in what I call the Hype Loop.

The Hype Loop is what happens when your product only makes sense in the context of “what’s next”.

You’re building around the assumption that the next model drop will fix what doesn’t quite work. That customers will care more, once they care more about AI. That their interest in you is just a lagging indicator of the overall hype curve. That your edge is being first, not being right.

It shows up in products that:

• Feel like demos, not useable tools

• Depend on screenshots, not usage metrics

• Constantly need to “switch models” because none of them quite solve it

These products surf the wave of what’s new – and crash the moment the wave flattens.

(Which is why we’re already counting 193 tools in the AI graveyard this year – and why the list of AI tools I’m using every day is still surprisingly short.)

The Freeze Test is how you escape that loop. If your product holds its value even when progress pauses, you’re not chasing hype. You’re building something that lasts.

What passing the Freeze Test looks like

We’ve been here before.

Let’s rewind to 1913. Henry Ford introduces the moving assembly line and cuts the time to build a car from 12 hours to just 90 minutes.

Suddenly, cars are no longer luxury items — they’re mass market.
Ford didn’t just “improve efficiency.” He reshaped the entire industry.

Of course, progress didn’t stop there.
Toyota came along years later with lean manufacturing and took the whole idea to another level.

But even without those later improvements, Ford’s original system still would’ve changed everything.

It didn’t need perfection.
It just needed to work.

That’s the Freeze Test in action.

The value of the assembly line didn’t depend on future breakthroughs — it held its own, right from version one.

Now imagine being a carmaker in 1914 saying:
“We’ll wait until someone invents a better version. This doesn’t seem perfect yet.”

You’d be out of business.

AI is the modern assembly line for intelligence.

It gives us cheaper, faster, scalable decision-making.
Just like Ford gave us cheaper, faster, scalable production.

Yet some teams are still waiting.
Waiting for the next model. The next tool. The perfect fit.

But you don’t need perfect.
You just need useful.

Build something with what we have today.
Build something that benefits from, but doesn’t require, whatever comes out tomorrow.

That’s the whole point of the Freeze Test.

Why this matters now

The pace of AI progress has trained us to expect constant upgrades.
Better models. Cheaper inference. New features every few weeks.

But that pace won’t last forever.
At some point, we’ll hit a plateau — even if just temporarily.

And when we do, the products that survive won’t be the ones with the flashiest demos.
They’ll be the ones that actually solve something.
That earn their place in a workflow.
That don’t break when the hype cools down.

We’re past the point where AI is just a curiosity.
The market is already shifting:

• From novelty → utility

• From demos → deployments

• From chasing hype → solving problems

We’re getting smarter about what sticks.

But the gap is widening — between teams that are shipping durable value and those still waiting for the next wave to carry them forward.

The Freeze Test is how you stop chasing someone else’s roadmap and start defining your own.

Here’s what that looks like in practice:

• A chatbot that eliminates a 5-hour manual task — even if the model behind it never improves again.

• A simple AI workflow that automatically drafts personalized outreach emails – without managing a gazillion templates.

• A document parser that just works for the formats your ops team handles every day.

• An internal agent that answers repetitive vendor questions — and keeps doing it without a model switch every quarter.

These aren’t moonshots. They’re grounded. Repeatable. Useful.
And they still make sense even if AI takes a break for a year.

One question to leave you with

Take the AI use case you’re currently working on — or thinking about.

Ask yourself two things:

1. What happens to this use case if AI innovation accelerates?

2. What happens to it if AI innovation stops?

If your use case benefits from the first and doesn’t mind the second, you’re on the right track. Building something real that holds its own, even when the hype wears off.

Because that’s what lasting products do.
They earn their place.

See you next Friday,
Tobias

P.S. Next workshop’s coming up on Wednesday — 1 Prompt, 100 Emails.
We’ll build a simple AI flow that generates outreach emails at scale — ones that actually feel personal and convert (without losing your mind over templates). And yes, we’ll make sure it passes the Freeze Test, too. Save your spot →