LLenny's Podcast
← All frameworks
InnovationRaiza Martin (Senior Product Manager, AI @ Google Labs)

The Magic Test for Control Surfaces

Users ask for knobs and sliders; shipping them literally is how a magical product becomes an ordinary one

Difficulty
Moderate
Time to result
~weeks to results
Steps
5
Confidence
91%

The most requested next feature for Audio Overviews was control — knobs, sliders, text boxes. Martin drew the mock, looked at it, and stopped: the controls were technically what people asked for but they made the product feel like a normal tool rather than a magical one. The framework is a design gate applied between user request and shipped control.

Origin

Raiza Martin's decision on the NotebookLM Audio Overviews roadmap in late 2024, when the obvious next release was configurability and she deliberately delayed to redesign the control experience.

Core principles

  • 01The one-shot magic of 'press generate and be surprised' is a real asset that controls can destroy.
  • 02Feature requests are evidence of a need, not a specification for the solution.
  • 03The gate question is 'does this feel like the same thing we shipped?' — control must inherit the product's magic, not sit beside it.
  • 04It is legitimate to take extra time on the control surface rather than ship the obvious version fast.
  • 05Not knowing what happens when you press the button is itself part of the value.

How to run it

  1. 1

    Collect the literal requests

    Listen to what users actually say they want. For Audio Overviews, users wanted to steer depth, tone and length — go deeper, be funnier, be more serious.

  2. 2

    Mock the literal solution

    Build the obvious version — the knobs, the sliders, the text boxes — and put it in front of yourself. The mock is the diagnostic instrument, not the deliverable.

    Pro tip Mock it fast and cheaply; its only job is to be looked at critically.

  3. 3

    Apply the magic gate

    Ask: does this still feel magical, and does it feel like the same product we shipped? Martin's verdict on the knobs mock was that it was 'great' but did not feel magical and almost did not feel like the same thing they had shipped so far.

    Pro tip Explicitly separate 'is this good?' from 'is this us?' — the knobs passed the first test and failed the second.

    Watch out 'It's what users asked for' is not a defence against failing this gate.

  4. 4

    Redesign the control experience to inherit the magic

    Rather than cancelling control, take deliberate time to make the control-and-steering experience itself magical and delightful — solving the underlying need (steerability) in a way that preserves the surprise and one-click ease of the original.

    Watch out Delaying control indefinitely is also a failure. The need is real; only the literal solution is rejected.

  5. 5

    Protect the surprise

    Preserve at least one path where the user gives minimal input, presses generate, and does not know what they will get. Martin identifies that not knowing what will happen is a core part of why people love the resume and check-in use cases.

    Pro tip Watch for the 'hyped up by two strangers' effect — the emotional payoff came from an output the user did not specify.

In the wild

The knobs that didn't ship

Facing consistent user demand for control, Martin's first instinct was to ship a bunch of knobs, sliders and text boxes. On seeing the mock she judged it good but not magical, and not recognisably the same product, so she took time to rethink the control experience instead.

The obvious configurability release was deliberately delayed in favour of designing a control experience that preserved the product's magic.

The resume upload

Users uploaded resumes and quarterly check-in notes with no configuration at all and got two AI hosts enthusiastically discussing their work. Googlers Martin had never met pinged her to say it boosted their confidence before meetings.

Martin identified the unspecified, one-click surprise as the source of the delight — a property any control surface must not destroy.

Common mistakes

Treating feature requests as specifications

Users asked for knobs and sliders because those are the controls they have seen before. Shipping them literally solves the stated request while eroding the property that made the product remarkable.

Confusing 'good' with 'on-brand'

Martin's own reaction to the mock was 'oh it's great' followed immediately by 'this doesn't feel magical'. A design can clear the quality bar and still fail the identity bar.

Removing all surprise in the name of control

The delight in the resume and check-in use cases came precisely from users not knowing what would come back. Full determinism would have killed the emotional payload that drove the sharing.

Is it for you?

Best for

Product leaders of AI products whose early magic came from a one-shot, opinionated output and who are now facing loud demand for configurability

Not ideal for

Professional or enterprise tools where deterministic, precise control is the core value proposition and surprise is a defect

From the transcript

the first thing that I thought was let's ship a bunch of knobs right like to me it was like what I was hearing people…

29:30

I'm actually taking a little bit of time to think about how do we make even that and control experience much more magical and delightful

30:00

imagine you click that and I think there's something really special about you don't know what what's going to happen right you just click generate

18:30

From the episode

Behind the product: NotebookLM

Raiza Martin (Senior Product Manager, AI @ Google Labs)