Zion

Writing about the psychology of money.
Why most investors underperform - and what actually works.

[Zion @Zi01nn]

@DaniilBuilds the payoff still worked because the bet was senior to the mistake

[Zion @Zi01nn]

@betedg When they disappear, the debt remains

[Zion @Zi01nn]

@DaniilBuilds people learn this lesson after the write off

[Edge @betedg]

Blackstone lost 100% of its original equity investment in one of its earliest deals.

The company was Edgcomb, a steel distributor Blackstone invested in in 1989.

At first, the business looked profitable.

But part of those earnings was tied to a powerful tailwind: rising steel prices increased the value of inventory Edgcomb already held.

That can make a distributor look more profitable without proving that its underlying economics have permanently improved.

Then steel prices turned.

The inventory gains disappeared. Earnings weakened. But the debt used to finance the business did not disappear with them.

That is where leverage becomes dangerous.

It doesn't just magnify a bad outcome.

It tests whether the earnings supporting the debt were actually durable.

Blackstone eventually lost 100% of its original equity investment.

The lesson from Edgcomb wasn't simply “use less leverage.”

It was more uncomfortable:

Before levering a business, make sure the earnings you're levering are real, repeatable and durable.

[Zion @Zi01nn]

@betedg interesting angle for this, saved it.

[Zion @Zi01nn]

@DaniilBuilds markets price the gap between expectation and delivery, the headline number is only the starting point

[Zion @Zi01nn]

@DaniilBuilds That’s why strategies that look great on paper quietly destroy real accounts over time

[Zion @Zi01nn]

@RaoulGMI The average person reaches their 50s looks at their pension and assumes they messed up
They didn’t mess up They followed the exact plan that was sold to them for decades
The problem is the plan itself stopped delivering Not the person who trusted it

[Zion @Zi01nn]

@HeyRohhit Most people watch this and still lose money
Not because the advice is wrong But because they can’t follow it when their emotions kick in
Buffett’s real edge was never the strategy It was the ability to sit still while everyone else panicked

[Zion @Zi01nn]

Most people collect tactics the same way they collect investing tips
They feel productive They feel like they’re improving
But without a clear strategy for their own behavior none of it compounds
The real bottleneck is almost never information It’s the inability to stick to a simple plan when emotions kick in

[Zion @Zi01nn]

Most multi-agent systems look clean in the diagram

Decentralized network
Hierarchical tree
Nice arrows
Everyone nodding

Then the agents actually start talking to each other

Coordination complexity hits
Shared information gets overwritten

One agent overrides another
And the system slowly starts producing confident nonsense

This is the part almost nobody designs for

They obsess over the structure
They ignore the failure modes that only appear when multiple agents run at the same time

I’ve seen this kill more projects than bad models ever did

Where does coordination actually break in the systems you’re building?

[Zion @Zi01nn]

@0xTaynew Imagine how he will cook for me🤝

[Zion @Zi01nn]

Most people are still building agent graphs and loops like it’s 2024.

They obsess over the architecture.
The routing.
The memory layer.

And completely skip the one thing that actually keeps the system alive:
a self-improving eval agent.
Anthropic already paid two years for this lesson.

Without a loop that constantly turns failed runs into new test cases, your beautiful graph just slowly starts lying to itself.

The agents keep running.
The scores look fine.

Until one day nothing works and nobody knows why.

This is where most multi-agent systems quietly die.
Are you still shipping graphs without a living eval loop?

[MIKE @mikenevermiss]

https://t.co/W4sHoE0ew0

[Zion @Zi01nn]

Most people are still building agent graphs and loops like it’s 2024.

They obsess over the architecture.
The routing.
The memory layer.

And completely skip the one thing that actually keeps the system alive:
a self-improving eval agent.
Anthropic already paid two years for this lesson.

Without a loop that constantly turns failed runs into new test cases, your beautiful graph just slowly starts lying to itself.

The agents keep running.
The scores look fine.

Until one day nothing works and nobody knows why.

This is where most multi-agent systems quietly die.
Are you still shipping graphs without a living eval loop?

[MIKE @mikenevermiss]

https://t.co/W4sHoE0ew0

[Zion @Zi01nn]

@0xRozzy Great post 🫡

[Zion @Zi01nn]

@patrik_agentsAI Incredible work! Gas

[Zion @Zi01nn]

@0xRozzy Interesting post brooo

[Zion @Zi01nn]

@rohanpaul_ai @alexandr_wang this is a real unlock people miss

[Zion @Zi01nn]

@patrik_agentsAI wild story

[Zion @Zi01nn]

@hanakoxbt people never use this for real...