Marius Horga

Tech Leader | Author | Entrepreneur

[Marius Horga @gpu3d]

🧵 I built a thinking OS to help you stop overthinking and start acting.

It’s called Think In Loops — and for the first time, it’s available to the public.

Here’s what’s inside and why it matters👇 1/6

[Marius Horga @gpu3d]

This book sale taught me a painful lesson about early optimization.

I saw a promotion and bought 14 books for $140. It was a deal I couldn't refuse. 🙃

Just days later, I realized I only needed two of them. If I had waited and bought just those two at full price, I would’ve spent $40 and saved $100 on books I will never read and can't return.

Andrew Ng said:
“Before optimizing, make sure you’re optimizing the right thing.”

That's what I should have done too, explore first. Optimizing too early just locks in the wrong decision.

[Marius Horga @gpu3d]

I still hesitate sometimes when I talk about my experience and it's not because I don’t know what I’ve done, but because my brain feels insecure.

David Rock said that uncertainty is experienced by the brain as a threat. That had influenced the way I behave now.

[Marius Horga @gpu3d]

I kept asking “what happened?” instead of “what’s been shaping this?”

I have meetings with my team almost every day, and for a while I noticed that the meeting would end and people felt frustrated.

I kept focusing on the words that were said but I missed what was causing the frustration: assumptions, unspoken expectations, and what people thought I already believed.

What we see is a conversation. What drives it lives beyond spoken words. That’s the Iceberg Model in action.

[Marius Horga @gpu3d]

This feels a bit nerve-racking to post, but here goes.

I’ve started offering coaching for first-time managers and leaders in tech.

I’m offering 10 free 30-minute strategy calls to start.

In these calls, I mostly listen and try to understand what you’re dealing with at work and where your time and energy are going.

If this sounds useful, DM me or share with someone you think would be interested.

[Marius Horga @gpu3d]

Leadership doesn’t start with a title.

I’ve seen people stepping up from every seat in an organization.

Leadership shows up where people choose to act, not where authority sits.

[Marius Horga @gpu3d]

I’ve seen this pattern more than once at work. A slow step shows up in a workflow, so we add AI there. That step gets faster but almost nothing changes overall.

Andrew Ng said it well: big gains come from redesigning workflows end to end, not from optimizing individual steps.

AI doesn’t create leverage on its own. Workflow redesign does.

[Marius Horga @gpu3d]

For a long time, I thought being helpful as a leader meant having an answer ready.

What I didn’t notice was how often that impulse made the room smaller. People just waited and held back on their own ideas.

When I stopped jumping in and let the space stay open a bit longer, the room changed. People started engaging, and conversations stopped routing back through me.

Leadership is less about answers and more about making room for everyone to contribute.

[Marius Horga @gpu3d]

At some point my calendar filled up without me noticing when it happened.

Each meeting made sense on its own, so none of them felt like a problem. It was only when I looked at the week as a whole that I realized there was no real space left to think or focus.

In systems thinking, this is called Emergence. Systems reveal outcomes that don’t exist in its elements taken separately.

[Marius Horga @gpu3d]

I saw this in the same meeting. Jane chose to speak even though she didn’t have to. She stepped outside her comfort zone on her own terms. Later, John was asked to speak so he kept it short and safe.

Both reactions made sense. When you choose the risk, the brain stays open. When the risk chooses you, the brain goes for safety.

What we call confidence or hesitation is often just the brain responding to choice versus surprise.

[Marius Horga @gpu3d]

I used to think progress on classical guitar meant improving speed and finishing the piece sooner.

What actually helped was slowing down, going with only a few measures at a time, and removing tension.

That’s a systems lesson I keep seeing everywhere: pain points tell you where the system suffers and leverage points tell you what can help the system heal but they are often times not easy to spot.

[Marius Horga @gpu3d]

I almost lost a relationship with a direct report without realizing it.

In our one-on-ones, he kept pushing me to see another side of how I was leading. Those conversations weren’t easy, but they were honest.

So one day I decided to open up and acknowledge that publicly and give him credit in front of the team.

It was a transformative experience for me as our trust reset, communication opened up, and his productivity even improved soon after.

Leadership shifted for me when I stopped protecting my position and found the courage to open up.

[Marius Horga @gpu3d]

Silence makes people uneasy, especially in groups.

Ask a question in a meeting and if no one answers right away, the silence is usually broken by the same person, often the leader or the loudest voice, not because they’re ready but because the quiet feels uncomfortable.

Most of the time, silence isn’t disengagement but thinking in progress. We just need to learn to live with it.

[Marius Horga @gpu3d]

At work, I kept asking an AI tool to predict which research jobs would fail overnight.

The predictions were decent but nothing changed. The real problem wasn’t knowing what might break. It was that no one acted on the signal.

Once we tied the prediction to a simple action, such as "flag the job early and adjust resources", failures dropped fast.

Prediction feels "smart" but it's action what actually moves things.

[Marius Horga @gpu3d]

Remember the situation from the other day: Jane said she couldn’t make a meeting after all and John assumed she’d backed out under pressure.

Later, John learned Jane had a personal family issue she couldn’t share at work. His story wasn’t wrong because it was careless but because it was incomplete.

Leadership changes when you stop treating first impressions as facts and start leaving room for missing information.

[Marius Horga @gpu3d]

Continuing on what I wrote yesterday. Jane missed a meeting. John assumed she caved under stress. Later, he learned she had a family matter she couldn’t share.

Delays is a systems thinking concept. The issue wasn’t the assumption but the delay before the correction. By then, behavior had already changed. Delays keep misunderstandings alive.

[Marius Horga @gpu3d]

Yesterday, Jane said she couldn’t attend a meeting. John assumed she caved under stress.

Today’s follow-up. We added a small AI step that separated facts from assumptions. So what did we learn?

Facts: Jane accepted, then declined.
Assumptions: stress, avoidance, disengagement.

Nothing else changed except the conversation that followed. That’s real AI leverage: not asking for answers, but for making assumptions visible.

[Marius Horga @gpu3d]

This scenario happened at my workplace. Jane said she’d join a meeting. The next day, she sent a short message saying she couldn’t make it. John immediately assumed she caved under stress and made an excuse. Later, it turned out Jane had a family matter she couldn’t share at work.

The message never changed. Only the story did.

This is what Chris Argyris calls the Ladder of Inference.

[Marius Horga @gpu3d]

Progress often slows even when effort stays the same.

That’s not failure.
It’s a balancing loop.

Systems push back to maintain stability. Early gains level off as the system adapts.

Reinforcing loops speed things up.
Balancing loops slow them down.
Both are always at work.

[Marius Horga @gpu3d]

People don’t pay as much attention to what leaders say as they do to what they react to.

Some examples are: what gets a quick response, what gets questioned, what gets ignored.

Those small reactions send people signals about what leaders think matters.

[Marius Horga @gpu3d]

We improved a triage model on one of our internal platforms. Accuracy went up and the metrics looked great but the workflow didn’t change.

The predictions were better but the system behaved exactly the same because accuracy is local but impact is systemic.

Andrew Ng said once that the hard part of AI isn’t crafting the best model but getting it to work in a system that creates value.