“This is the worst AI model you’ll ever use.”

A quote from Kevin Weil, CPO of OpenAI, that really stuck with me!

I’ve been working with AI tools almost daily, testing, designing, and shipping.
Every time a new drop lands — like OpenAI’s memory and Dropbox integration yesterday, or DALL-E’s big update two weeks ago — I realise something:

We’re building with tools that evolve faster than our thinking.

This is the new normal.

The teams that thrive aren’t following last month’s roadmap.
They’re building on the edge of what’s possible today and adapting fast when it shifts.

But I still catch myself stuck in old habits.
And I see teams around me doing the same.
We’re building like it’s 2020.
Same assumptions.
Same workflows.
Same patterns.

Meanwhile:

The tools have changed
The risks have changed
And our responsibility has changed

You can’t just “add AI.”
You have to rethink everything:

Where the data comes from
What the model understands
When it’s right — and when it’s wrong

Finding data is AI’s strength.
Validating it? That’s on us.

Yet I rarely see teams:

Writing evals
Checking source reliability
Building the right guardrails

That gap is real. And risky.

The design side has its own issues.
Most AI tools don’t say, “I don’t know.”
They speak with confidence — even when hallucinating.

It reminds me of Forbidden Planet, 1953 film about new worlds and boundless potential.
Unchecked power. Flawed reasoning.
The outcome? Not great.

If AI can’t admit doubt, show its logic, or signal uncertainty, it can’t build trust.
And without that, we’re just wrapping UI around guesswork.

AI isn’t a side project.
It’s a product. A collaborator. A risk.

So here’s my question:
Are we designing with the model?
Or are we letting it drive blind?

Has your team adapted to this new normal?
Would love to hear what’s working.