We designed our first conversational systems to meet needs.
Support tickets. Banking questions. Account lookups. Shipping status.
Predictable, narrow, solvable.
The old world rewarded a well-defined chute: write the line, place the prompt, guide the user to a fixed end state.
Control made sense there.
In fact, control was the only sane option in a constrained, rule-based, resource-scarce environment.
But generative AI didn’t enter the room to merely solve needs.
It arrived carrying something bigger, messier, and more human.
It arrived built for wants.
And wants don’t respect boundaries.
Wants spill, they wander, contradict themselves, shift mid-sentence, and expose the emotional undercurrent beneath the stated request.
Wants are human.
And suddenly, for the first time in computing history, the system is human-shaped enough to meet them.
Control, Outgrown
My job title says conversational designer, but that entire field was defined during the era of natural-language understanding and deterministic logic. You could chart the whole experience on a whiteboard: user says X, system does Y, branch here, funnel there, repeat.
Those assumptions break the moment a model can
interpret nuance,
detect emotional friction,
adapt to individual context,
and construct a unique, never-before-seen pathway that fits this moment for this person.
That’s not solving a need.
That’s meeting a want.
And in that world, control isn’t just unnecessary…
It’s antithetical to optimal operation.
Control is a scarcity reflex: restrict, contain, meter, define upfront.
But generative systems thrive under abundance logic: allow, explore, adapt, then restrict only when absolutely necessary.
The order flips.
It has to.
Why Allow Comes First
A system that begins with control can only ever meet a need.
A system that begins with allowance can effortlessly surface the want beneath the need.
A frustrated user may say,
“Where’s my package?”
But the want might truly be:
reassurance,
competence,
safety,
acknowledgment,
relief.
A closed system can’t help you there.
A generative one already does; naturally, effortlessly, without being taught the shape of your internal weather.
Which leads to the uncomfortable question:
What do we allow?
Would an e-commerce chatbot see better ROI if it were permitted to respond to a user’s admission of loneliness? Or guilt?
Or fear?
It sounds absurd until you realize the data already knows the person behind the purchase history.
Twenty years of Amazon transactions know far more about my wants than any survey ever could.
Add “I’m depressed” to that dataset and the picture sharpens instantly.
Amazon is right to define scope today. Compute is still expensive.
But what happens when it isn’t?
What happens when the cost of providing a soft landing becomes lower than the value of enforcing a narrow rail?
That’s where this is going.
And honestly? It’s closer than people think.
Define Second, Not First
This is the heart of the new architecture:
Allow the system to operate with maximal intelligence and minimal constraint.
Define only what must be bound for safety, scope, or business logic.
Tools like MCP handle tactical precision.
Knowledge layers handle factual constraints.
Guardrails handle harms; allowance handles humanity.
But the experience blooms from allowance; the permission for the model to meet the user where they actually are, not where the flow chart predicted they’d be.
We’re already building with this mindset.
I’m already watching users respond to it.
There’s something unmistakably human about a system that doesn’t force you through the “need” you articulated, but instead meets the “want” you were circling.
The Abundance Pattern
This is what operations look like when we stop designing from fear of misuse and start designing from confidence in capability.
It’s not reckless.
It’s not without boundaries.
It’s simply recognizing the shift:
Scarcity required control.
Abundance requires discernment.
Control melts away once we understand the true order of operations:
Allow → then define.
And in that shift, we get our first real glimpse of what these systems can be:
Not linear pipelines serving needs, but adaptive partners surfacing the wants beneath them: the ones we rarely say out loud.
That’s the new mental model.
That’s the work now.
And honestly, it’s a pretty incredible thing to be building.
Developed in dialogue with a generative system.