The carriage problem in AI product design
Early automobiles looked like carriages with engines bolted on.
The engine was new.
The structure was old.
Real transformation did not happen when the horse was replaced.
It happened when the vehicle was redesigned.
AI is at the same stage.
The bolt-on instinct
Across the tech landscape, AI is being added to existing products.
An assistant panel.
A summarise button.
A recommendation layer.
This is not redesign.
It is augmentation.
The product stays the same.
The workflow stays the same.
The architecture stays the same.
AI becomes an enhancement, not a foundation.
That limits its impact.
Why bolt-on AI underperforms
Adding AI to legacy systems creates friction.
User experience feels tacked on.
Capabilities are constrained by old workflows.
Interfaces were not designed for generative behaviour.
You get incremental improvement.
You miss structural change.
This is not failure.
It is a design ceiling.
From smarter features to new categories
AI is not a smarter feature layer.
It is a new medium.
Electricity did not make candles brighter.
It created electric grids.
The internet did not optimise fax machines.
It created digital platforms.
Mobile did not refine desktops.
It created new behaviour.
AI will not simply make products smarter.
It will create different products.
The Netflix thought experiment
Consider Netflix.
Today it analyses viewing data, segments audiences, and commissions content designed to perform at scale. That is a broadcast model.
Now consider advanced generative systems for video, voice, and sound. Tools such as OpenAI’s Sora, Google’s Veo3, and Flow are early signals of this direction.
Instead of recommending content, Netflix could generate it.
A story shaped by your viewing history.
A plot tuned to your genre bias.
An ending aligned with your engagement patterns.
Not branching narratives.
Dynamic generation.
The unit shifts from mass production to personal synthesis.
That is not optimisation.
It is category redesign.
Broadcast versus narrowcast
Broadcast creates one version for millions.
Narrowcast creates millions of versions for individuals.
The first model scales distribution.
The second model scales generation.
AI enables the second.
That changes economics, licensing, and creative ownership.
It also changes product design.
The design shift AI requires
Most teams ask:
How can AI improve our product?
That framing keeps the architecture intact.
The better question is:
What becomes possible if AI is the product?
That reframes everything.
Data structures change.
Interfaces change.
Feedback loops change.
You move from static releases to evolving systems.
Designing from first principles
AI-native design starts with capability, not UI.
What can be generated?
What can be personalised?
What can adapt in real time?
Then structure the product around those properties.
Not around menus.
Not around feature lists.
Around dynamic systems.
Static products versus generative systems
Traditional software ships fixed functionality.
AI systems generate functionality.
Traditional releases are versioned.
AI systems are continuously adapting.
Traditional experiences are identical across users.
AI experiences diverge by context.
If you design generative systems like static software, you waste leverage.
What this means for product leaders
Do not ask where AI fits in your roadmap.
Ask whether your roadmap assumes the wrong model.
If AI is bolted on, the architecture remains legacy.
If AI is foundational, the product behaves differently.
The companies that win will not ship the most AI features.
They will redesign the operating model.
And redesigning the operating model is where the leverage lives.