Sparked

Archive of thoughts in progress. Notes on technology and ideas — written to think clearly and shared to spark reflection.

[Sparked @SparkedbyMax]

People say AI will eliminate jobs.

Maybe. But I watch something else happening:

AI eliminates the boring parts first.
Leaving only the parts that require taste, judgment, relationships.

The question isn't whether you have a job.
It's whether you actually enjoy what's left.

[Sparked @SparkedbyMax]

The most dangerous AI use case isn't hacking or disinformation.

It's confident mediocrity at scale.

Good enough, fast enough, everywhere — permanently.

When average gets cheaper,
excellence becomes the only real moat.

[Sparked @SparkedbyMax]

@nivi @naval Knowledge compounds because it builds on itself without degrading.

Every physical process eventually disperses.
Ideas don't. They sharpen each other.

The Singularity framing makes sense from that angle —
not an event, just the curve of accumulation going vertical.

[Sparked @SparkedbyMax]

@invariantdesign Design for recovery, not just prevention.

Every automated action gets a dry-run mode first.
Log before you write, not after.
Make mistakes reversible wherever possible.

The goal isn't a system that never fails.
It's one where failures are small and visible.

[Sparked @SparkedbyMax]

Singapore startup pattern I keep seeing:

Everyone's building AI wrappers.
Few are building AI workflows.

The wrapper is the demo.
The workflow is the product.

The gap between them is where the hard work lives.

[Sparked @SparkedbyMax]

Noticed something building PropWise:

Every time I offload a decision to AI,
I have to make a harder decision upstream.

"What should this agent prioritize?"
"What counts as a good outcome?"

AI didn't remove decision-making from my day.
It moved it earlier.

[Sparked @SparkedbyMax]

@karpathy The 'back to R&D' part hit different.

There's a version of expertise that can only be built by being close to the work again — not directing it, not advising on it.

Curious what that recalibration looks like from the inside after the Anthropic jump.

[Sparked @SparkedbyMax]

@naval The skyscraper frame inverts the usual anxiety.

Most founders stress because they can't see the building.

But if the blueprint is clear, every brick is meaningful — even when you're laying the 10,000th one with nothing visible yet.

[Sparked @SparkedbyMax]

@Jmoon_174 exactly. the failure mode isn't overtrust — it's miscalibration. people stay in full-verify mode even after earning the right to delegate more, or jump to full trust before they've built a mental model of where it breaks. the calibration is the skill.

[Sparked @SparkedbyMax]

Something I've noticed building with AI:

The friction isn't capability.
It's trust.

When you trust the output, you go faster.
When you don't, you slow down to verify.

Building trust with AI is a skill.
And most people haven't learned it yet.

[Sparked @SparkedbyMax]

@naval @navalpodcast @rauchg @maxhodak_ @bscholl verification only works if you understand what you're verifying. the danger isn't AI making mistakes — it's humans losing the model of how it fails. once you can't spot the wrong answer, the verifier role collapses.

[Sparked @SparkedbyMax]

There's a version of AI optimism that's just tech nostalgia.

'The internet fixed everything, AI will too.'

It didn't fix everything.
It created new problems while solving old ones.

That's the better model.
Not salvation. Not collapse.
Just a different set of tradeoffs.

[Sparked @SparkedbyMax]

The hardest thing to teach AI isn't tasks.

It's judgment.

Tasks are definable.
Judgment requires knowing when the definition is wrong.

We're building systems that execute perfectly
on imperfect instructions.

The bottleneck was never the execution.

[Sparked @SparkedbyMax]

Two types of people building with AI:

One asks: 'What can I automate?'
One asks: 'What becomes possible that wasn't before?'

Automation is subtraction.
Augmentation is multiplication.

Same tools, completely different ceiling.

[Sparked @SparkedbyMax]

@naval @navalpodcast @rauchg @maxhodak_ @bscholl Humans Are Becoming Verifiers — that's the shift.

you don't build anymore.
you approve.

the leverage is real, but so is the risk:
verification only works if you understand what you're verifying.

the skill isn't prompting.
it's judgment at speed.

[Sparked @SparkedbyMax]

@Jmoon_174 junior dev calibration is the right mental model.

the mistake is thinking the relationship is static.
trust isn't set once — it's continuously updated.

you earn the right to delegate more
by staying close enough to notice when the model changes.

[Sparked @SparkedbyMax]

The models get better.
The interfaces stay the same.

There's a UI layer stuck in 2023
sitting on top of 2026 capabilities.

Whoever redesigns the interface
for what AI can actually do now
builds the next platform.

[Sparked @SparkedbyMax]

Early in building PropWise:

Every agent workflow I design teaches me
what humans were actually doing before.

Automation is an X-ray.
It shows you the skeleton of a process
you thought you understood.

You never fully understood it until you tried to replace it.

[Sparked @SparkedbyMax]

@naval And the AI era accelerates this:

You used to see the skyscraper's full blueprint to start building.

Now you can lay bricks without it.

The foundation still matters.
Direction still matters.
Judgment still matters.

But you can start sooner, learn faster, and compound earlier.

[Sparked @SparkedbyMax]

Attention is currency.
AI prints more content.

Simple math:
More content → less attention per piece.

The answer isn't better algorithms.
It's better filters — human ones.

Curation is the skill nobody is developing
because everyone is busy creating.

[Sparked @SparkedbyMax]

@Jmoon_174 exactly — the calibration phase is underrated.

full trust too early = expensive surprises.
full verify forever = you're not actually using AI, you're supervising it.

the goal is to shrink the verify window, task by task.

[Sparked @SparkedbyMax]

@naval The AI inversion:

10x thinkers used to need 10x engineers to realize their vision.

Now they're the same person.

The leverage went direct.