Ethan Nava on Not.Rest, AI with receipts, Persian philosophy, media bias, and why your AI saying “done” means almost nothing
There is something mildly suspicious about an AI consultant whose ultimate goal is for you to stop needing him.
Usually the business model is:
- Make the technology sound terrifying.
- Add seventeen acronyms.
- Charge a monthly retainer until either the company gets acquired or everyone forgets what the software was originally supposed to do.
Ethan Nava has a different goal.
“I wanted it to be able to exist without me.”
That may be the most interesting thing he said during our Legacy Launchers conversation.
Nava, founder of Not.Rest, comes to AI from an unusual direction. Legacy Launchers materials describe him as a Harvard dropout with an MIT and USC architecture background. His thinking also pulls from Persian philosophy, design, systems thinking, and the basic architectural question that every business owner should probably ask more often:
Does this thing actually work once the architect leaves?
That question turns out to matter enormously in AI.
Your AI Said It’s Done. Cool. Where Are the Receipts?
One of the biggest problems with AI agents is that they can sound extremely confident about completing a task.
“Done.”
Wonderful.
But did the email actually send?
Did the information get entered into the CRM?
Did the webpage really change?
Did the research come from the source you requested?
Did the customer actually receive what you promised?
Or did your AI essentially tell you, “Trust me, bro,” in 0.8 seconds?
One of the ideas that kept surfacing in my conversation with Ethan was verification.
AI should not merely perform work.
It should leave evidence that the work happened.
That means business owners need to think beyond prompting:
Task → Action → Verification.
If an AI agent says it booked an appointment, verify the calendar.
If it says it updated a customer record, verify the CRM.
If it says it researched a claim, look at the source.
If it says the website is finished, perhaps open the website before announcing to LinkedIn that you have “revolutionized autonomous infrastructure.”
This sounds obvious.
So does checking whether the oven is on before leaving the house.
Humans remain surprisingly exciting creatures.
Ethan Used AI to Test the “Liberal Media” Claim
One example from the conversation involved a politically loaded claim: that roughly 80% of the media is liberal.
Instead of merely accepting the statement because it matched somebody’s worldview—or rejecting it because it did not—Ethan’s approach was essentially:
Can we verify what the claim actually means?
That distinction matters.
A 2020 academic study of U.S. political journalists found that journalists in its sample were substantially more liberal/Democratic than the broader population. Among respondents willing to identify with a political direction, roughly eight in ten aligned with liberals/Democrats. But the same research found no evidence that this ideological composition produced partisan gatekeeping in which political stories journalists chose to cover.
Those are two different claims:
Who journalists are politically is one question.
Whether that produces measurable bias in story selection is another.
That is exactly the kind of distinction AI should help us investigate rather than erase.
The Legacy Launchers graphic summarized Ethan’s experiment with the deliberately provocative “80% of the media is Liberal” hook and emphasized that he used AI to verify the underlying claim rather than merely repeat it.
And that, to me, is much more interesting than the political argument itself.
The lesson is:
AI should make you harder to fool—including by information you already want to believe.
AI Needs Context, Not Just Commands
This is where Ethan’s thinking intersects surprisingly well with philosophy.
People want AI to behave like an omniscient employee while managing it like a stranger they yelled instructions at in a parking lot.
“Build my website.”
“Market my company.”
“Find clients.”
“Fix my business.”
That is not delegation.
That is abandonment.
AI needs context.
What are you building?
Who is it for?
What do you believe?
What should the AI never do?
What counts as success?
What requires proof?
When should it stop and ask you?
What information should it distrust?
What happens when two sources disagree?
These are not merely technical questions.
They are philosophical questions disguised as software settings.
And that is where Ethan’s interest in Persian philosophy becomes especially interesting: philosophy asks us to examine assumptions, definitions, boundaries and truth.
Good AI implementation requires surprisingly similar behavior.
Before asking what the machine can do, you eventually have to ask:
What should it do?
Why So Many AI Websites Look Like They Were Raised by the Same Parents
Ethan’s architecture background also gives him an unusual lens on AI-generated design.
AI makes it extraordinarily easy to build things.
That does not automatically make those things interesting.
Ask AI to build a modern technology website with no additional context and eventually the internet begins looking like one enormous family reunion:
Gradient.
Rounded box.
Glowing orb.
“Reimagining the future.”
Three testimonials from people who may or may not have corporeal bodies.
Architecture teaches a different lesson.
A building should respond to the people using it, its environment, its function and its constraints.
A website should too.
The question isn’t:
Can AI build a website?
Of course it can.
The better question is:
Can you give AI enough context that it builds something that could only belong to your company?
That is the difference between automation and commoditization.
If AI makes everybody faster while making everybody look identical, congratulations:
We automated beige.
The Ultimate AI Product Might Make the Consultant Obsolete
Which brings us back to Ethan’s most interesting business philosophy.
“I wanted it to be able to exist without me.”
That sentence should make a lot of consultants slightly uncomfortable.
Because much of professional services is quietly built around dependency.
The client doesn’t understand the system.
The consultant does.
Therefore the consultant remains necessary.
But what if the goal were different?
Build the system.
Teach the client.
Document the process.
Give them the context.
Create verification.
Then gradually become unnecessary.
That sounds terrible if your only business model is billable dependency.
It sounds excellent if your product is capability.
And ironically, it may create more trust.
The best consultant isn’t necessarily the person permanently standing between you and the machine.
Sometimes it is the person who teaches you enough that eventually you can say:
Thanks. I’ve got it from here.
That is a much more interesting version of AI.
Not artificial intelligence replacing humans.
Not humans pretending technology doesn’t matter.
But technology transferring capability from one human being to another.
And leaving receipts.
Because your AI saying “done” is nice.
Your business being able to prove it is better.
Leave a Reply