Tony Fadell is the product person behind the iPod, iPhone, and Nest.

In this interview, what struck me the most wasn't those Apple stories, but a very simple judgment:
The truly difficult product decisions often can't wait for data to tell you.
Especially in v1.
V1 doesn't have enough users, no stable path, no clear benchmark, and even "what counts as good" hasn't been defined yet.
At this point, it's not about dashboards, but taste, opinion, and judgment.
This will become even more obvious in the AI era.
Because execution will become increasingly cheap. Writing code, generating pages, editing videos, making prototypes, running scripts—all will be pushed to very low costs by agents.
But the problem also reverses: what do you want it to do? What shouldn't it do? To what extent is it considered overdone? Where must it stop and wait for human judgment?
These won't automatically disappear as models get stronger; instead, they will become even more important.
I'm increasingly convinced that the core of personal AI workflows isn't "removing humans," but rather taking humans out of low-value execution and putting them back into taste, boundaries, and validation.
Agents handle throughput.
Humans decide what is worth processing.
That's also why I'm building my own workflow now, constantly adding SSOT, active-task, thread routing, and HTML gate.
On the surface, it looks like task management.
But essentially, it's answering a deeper question:
When the cost of execution approaches zero, is your judgment system strong enough?
source:
- Tony Fadell / Lenny's Podcast