The case for building AI based products differently

Authenticity isn’t performative. It’s not loud or algorithmically optimized. It’s courageous, clear, and deep. Brands that care about authenticity care about nuance. And nuance is HARD. Especially when the dominant trend is to bolt AI onto EVERYTHING, slap a wrapper on GPT, and call it a product.

We’ve seen it too many times. And to be blunt: it’s just NOT it!

A good technical solution, AI or not, MUST do two things:

1. Meet people where they are and,
2. Empower them to delegate with trust.

That takes thoughtful design, clear intent, and yes, CONSTRAINTS (lots of it).

Look at what Apple is doing with its AI architecture: placing AI in key moments where it saves just a bit of time, predictably, by delegating smartly. That’s mature planning and future thinking.

A good example of fitting AI into a product with grace? Text-to-speech or speech-to-text for translation. It’s invisible, seamless, and saves time in a way users can feel. That fits.

But building your entire app as a ChatGPT wrapper with some prompt engineering? Sure, it might help you test a POC. But it won’t scale into a REAL product. Not without rigor, thoughtfulness, and a whole lot of rewrite.

Chances are, if you’re building something that leans completely on a model trained on the internet’s loudest common voices, you’re not solving a real problem. You’re remixing the obvious. That’s not creative — it’s just lazy(in our opinion).

While AI excels at generalizing, it still struggles to specialize in any key area (at least today). Which is why the way we—along with many other bespoke teams around the world—approach AI differently by NOT chasing implementation for implementation sake. We believe in deeply embedding the intelligence in contextual, intentional, and always in service of action manner. Not text based abstraction.

Because AI is the spice, its NOT the whole soup and you don’t just serve a bowl of cumin.

Our 3 key principles for AI implementation are:

1. Actionable Insights, beyond just chatbots

Let’s be honest: all-encompassing chatbots are a TERRIBLE user experience(in our opinion). They’re unclassy, predictable, and lack the tact and structure that good design demands.

We avoid that trap by designing for schema-based actionable responses.

Our method?
We feed AI a limited, curated context:

• Where the user is and where they’re going.
• What we know about their preferences.
• How we want the system to respond—structured, modular, and/or chain-able.

For example, instead of prompting:

“User is asking how to make pasta. Reply in a list format.”

We prompt:

“What are the ingredients required to make {dish=pasta}? Include approximate time (in minutes), tools required. Return response in JSON.”

This gives us structured data. Not prose.
From there, we build UX layers:
Nearest grocery store. Estimated grocery size. Calendar suggestions.

Even chat interfaces can leverage this implementation that build from a structure, but shaped by it. It’s still AI, but refined, data driven and extremely practical. A lot of time, this can be done completely independently without requesting user for a prompt.

2. Healthy AI-economics (or what we call prompt-economics)

Here’s the thing: if your AI costs don’t scale sensibly with usage, you don’t have a product—you have a ticking cost bomb.

We believe AI-native products must be cost-aware by design. Not because we want to penny-pinch, but because healthy economics lead to better systems. That’s why we obsess over:

• Prompt and context window optimization
• Output reuse and caching
• Token budgeting and fallbacks

We’ve spent countless hours here. Not just for ourselves—but for clients who want predictable, investor-grade economics baked into their product DNA. It's not glamorous work, but it’s the kind of backbone good systems are built on.

Unicorns are cute, but we build horses that can run long distances.

3. Chain-able, predictable & repeatable UX

This is the one we love the most.

To us, design-engineering isn’t about visual polish. It’s about predictable, intelligent systems that serve users without shouting. Think Apple’s Dynamic Island—not AI-powered, but a brilliant example of showing up exactly when and where it matters.

We believe AI should do the same.

A well-architected UX doesn’t just feel good—it enables smart layering of data, actions, and recommendations.

When paired with structured insights and cost-efficient delivery, the experience becomes more than functional—it becomes trustworthy.

That’s where loyalty is built and that’s where growth begins.

We’ve said it before: AI isn’t the main character in a product.
It’s the quiet co-pilot in the background.
It saves seconds, not solves everything.

Because when you build with authenticity at the center,
you don’t follow trends—you build on truth.

And that makes all the difference.