I had one of those conversations today that leaves several ideas bouncing around your head long after it ends.
They looked unrelated at first: AI agents, Bending Spoons, European banks, corporate data leakage, founder psychology. But they all pointed toward the same conclusion:
Environment matters more
than the individual component inside it.
Sometimes more than the model. Sometimes more than the founder.
1. AI Models Are Becoming Commodities
Frontier models are increasingly starting to resemble electricity from a socket. Obviously, quality still differs. But the competitive advantage is moving upward in the stack.
The interesting question is becoming less: Which model are you using?
And more: What environment did you put it into?
The harness. Context. Permissions. Tools. Memory. Evaluation. Agent coordination. Feedback loops.
Connect a frontier model to a terrible engineering system and congratulations: you’ve invented a machine for producing terrible code much faster.
The model provides capability. The environment determines what happens to it.
2. Why Emotional Distance Can Be Worth Billions
Bending Spoons is a fascinating example of the same principle applied to companies. They acquire mature software businesses, centralize operations, cut costs, change pricing, and restructure organizations with a brutality that founders often simply cannot reproduce. There’s a reason.
A founder remembers employee number seven. They remember the customer who signed up in 2018. They remember building the first version at 3 a.m.
The new owner remembers none of this. Sometimes that’s an advantage.
Growth occasionally requires removing emotion from decisions, or handing the decision to someone who doesn’t carry the emotional history. Uncomfortable, but useful.
3. AI Agents Can Become External Discipline
One particularly useful pattern: Give an agent your strategy. Then let it stop you when you’re about to violate it.
We usually think of agents as machines for doing things. But perhaps one of their more valuable functions is preventing us from doing things.
Founders are exceptionally good at inventing rational explanations for why today’s exciting distraction is actually strategically essential.
An agent with persistent context can simply reply: “No. You explicitly said we weren’t doing this.” But there’s a catch.
AI can reinforce discipline. It cannot manufacture your willingness to change.
A founder who has outgrown the startup but hasn’t outgrown their own reflexes can deploy fifty agents and remain exactly where they were. Only faster.
4. Regulation Compounds
Regulation isn’t merely a cost. It’s a compounding dependency.
A European bank implementing something new doesn’t simply comply with today’s regulation. It must make today’s requirements coexist with yesterday’s requirements, internal controls, legacy architecture, audit procedures, and whatever arrives tomorrow.
Eventually enormous engineering capacity goes into maintaining regulatory compatibility rather than improving the product.
The interesting competitive question therefore isn’t simply who has better technology. It’s who can still move after accumulating ten years of constraints.
5. Corporate Data Leaks Through Humans
Companies worry about models stealing their data. The more immediate problem is employees enthusiastically giving it away.
Architecture. Customer information. Internal strategies. Source code. Microservice names. Paste. Enter. Done.
Banning AI tools doesn’t solve this because useful technologies have an annoying tendency to get used. A better approach is environmental again: build a filtering layer that detects and removes sensitive information before it reaches an external model. Don’t rely entirely on perfect human behavior.
Design the environment assuming humans will behave like humans.
The Steve Jobs Problem
Everyone wants to learn how to become Steve Jobs. There’s a statistical inconvenience here. History remembers the one who survived. There were countless obsessive, stubborn, unreasonable founders with absolute confidence in their vision.
Most of them simply became obsessive, stubborn, unreasonable founders of failed companies. We crowned the survivor and reverse-engineered a philosophy from him.
So I’m increasingly skeptical of formulas for genius. I’d rather build systems. Your tools. Your constraints. Your feedback loops. Your people. Your environment.
Talent matters. But the right environment compounds talent. And sometimes it carries you through the places where talent alone isn’t enough.
Build the environment first.
The environment does more than determine what a capability produces.
Over time, it also teaches the people inside it what to trust, what to attempt, what to escalate, and what to stop doing.
The environment shapes the output, but it also shapes the behavior that keeps producing it.