Signal detection: the skill behind great leaders (and great AI agents)
One underrated skill of great thinkers is not just reasoning. It is signal detection. The ability to see what actually matters before it becomes obvious.
What signal detection looks like in practice?
In 1997, Steve Jobs returned to an Apple bleeding from too many product lines. Over 350 SKUs. His first major decision was not to fix the products. It was to cut them. He reduced the lineup to four. Fewer signals to track, faster decisions, clearer identity.
This was not instinct. It was ruthless filtering. Jobs famously said that deciding what not to do is just as important as deciding what to do. He applied that to attention, not just strategy.
Or consider Jeff Bezos and the regret minimization framework. When deciding whether to leave his finance job to start Amazon, he asked himself which choice he would regret more at age 80. One question. One signal. Everything else was noise.
They did not process more. They filtered harder.
Why signals are hard to spot?
The problem is not lack of information. It is that visibility and importance are no longer correlated. The loudest signal in the room is rarely the most consequential one.
Early signals tend to be weak, incomplete, and uncertain. Noise, by contrast, is often loud, urgent, and emotionally compelling. A bad review, a competitor announcement, a some post going viral.
The most dangerous noise is not irrelevant. It is the almost relevant. The thing that looks like it matters, demands a response, and quietly pulls you off course.
Without a filter, your attention gets captured by whatever is most visible, not whatever is most important. The result is reactive decision making. You optimize details that do not move the needle while missing the shifts that do.
How AI agents actually do this?
Modern AI agents do not have intuition. Well designed ones enforce signal filtering through a reasoning loop.
At each step, the agent evaluates:
- Is this relevant to the current goal?
- Does this new data change my plan?
- What can I discard?
The agent does not try to process everything at once. It continuously compresses its context into what matters right now.
But here is what most people miss. A poorly designed agent does not fail because it reasons badly. It fails because it carries too much context forward. It accumulates information without asking what actually changed. That accumulated weight degrades every decision that follows.
The best agents and the best thinkers do not just ask what do I know. They ask what does this new information actually change. Intelligence alone does not solve this. The ability to drop the irrelevant at each step does.
Building your own filter
Most people ask themselves what should I do next. That is a planning question. It assumes you already know what matters. The more powerful questions come before it.
The three question filter
1. What actually matters here? Forces you to prioritize before acting.
2. What looks important but is not? Surfaces the noise masquerading as signal.
3. What would I need to see to change my mind? Keeps you open to real signals as they emerge.
The third question is the one almost everyone skips. It is the one that keeps signal detection alive over time instead of collapsing into confirmation bias once a decision has been made.
In practice, this might look like a product team receiving a loud feature request from a vocal user group.
Instead of acting on the volume, they run the filter.
Does this request change retention, revenue, or strategic positioning? If not, they treat it as noise for now.
But they don’t ignore it completely. They define what would change their mind:
If multiple customers request this within the next quarter, or if churn can be linked to this missing feature, they will revisit the decision.
The real skill
Attention is finite. The leaders and systems that perform best over time are not the ones that take in the most information. They are the ones that continuously decide what to ignore.
Better thinking is not just better reasoning. It is better filtering.
The real advantage is not seeing more. It is learning what to ignore.