“Measure, adjust, repeat”
Feedback Loop
Explain it like I'm five
Imagine you're trying to fill a bathtub to exactly the right temperature. You turn on the hot water, stick your hand in, and it's too cold — so you turn the hot up. Now it's too hot — so you turn it down a little. You keep checking and adjusting until it's just right. That check-and-adjust loop is a feedback loop. Your body does the same thing with temperature, blood sugar, and balance — constantly measuring and correcting, thousands of times a second, without you even thinking about it.
The Story
In 1788, James Watt attached a pair of spinning metal balls to a steam engine. As the engine sped up, centrifugal force pushed the balls outward, which partially closed the steam valve, which slowed the engine down. As it slowed, the balls dropped, opening the valve again. This "governor" was the first precision-engineered feedback loop — but the principle was ancient. Windmill fantails had been doing it for decades, and water clocks in Alexandria used float valves two thousand years earlier. Watt just made the loop fast enough to matter for industrial machinery.
The real breakthrough came when engineers realized the same math governed all of them. Norbert Wiener's cybernetics in the 1940s showed that a thermostat, a human reaching for a cup of coffee, and an anti-aircraft targeting system all share the same structure: sense, compare, correct. Biologists recognized that hormonal regulation — insulin responding to blood sugar, cortisol responding to stress — was the same loop running in flesh. Economists saw it in central banks adjusting interest rates to target inflation. The feedback loop wasn't an invention; it was a discovery of something that already permeated every self-regulating system in nature.
The frontier is everywhere the loop is still open. Urban traffic systems still run on fixed-timer signals while real-time sensor data goes unused. Most classrooms deliver content at a fixed pace regardless of whether students are keeping up or falling behind. Precision agriculture is beginning to close the loop — soil moisture sensors driving irrigation valves in real time — but most farms still water on a schedule. Wherever you find a system running on a timer instead of a sensor, you've found an open loop waiting to be closed.
Cross-Domain Flow
Technical Details
Problem
How do you keep a system at a desired state when external conditions constantly change?
Solution
Continuously measure the system's output, compare it to the desired state, and feed the difference back as an adjustment to the input.
Key Properties
- Sensor — measures current state
- Comparator — computes error between actual and desired
- Actuator — adjusts input based on error
- Delay — the lag between action and effect determines stability
Domain Instances
Thermostat / PID Controller
Control EngineeringThe PID (proportional-integral-derivative) controller is the workhorse of industrial automation. It continuously calculates an error value — the difference between a measured process variable and a desired setpoint — and applies a correction based on proportional, integral, and derivative terms. From cruise control to chemical plants, PID loops regulate billions of processes worldwide.
Key Insight
The three terms of PID — react to the present error, accumulate past errors, and anticipate future errors — are the minimal complete strategy for any feedback system.
Homeostasis
BiologyEvery living organism maintains internal stability through feedback loops. Blood sugar rises after a meal; the pancreas releases insulin to bring it down. Body temperature rises; sweat glands activate. The hypothalamus acts as the comparator, hormones as the signal, and organs as the actuators. These loops run continuously, involuntarily, and have been refined by 3.8 billion years of evolution.
Key Insight
Biology doesn't just use feedback loops — biological organisms ARE feedback loops, nested thousands deep. Life is what happens when enough loops close.
Central Bank Interest Rate Policy
EconomicsCentral banks measure inflation and employment (sensor), compare them to targets (comparator), and adjust interest rates (actuator). The feedback loop is explicit: the Federal Reserve's "dual mandate" is literally a setpoint, and rate decisions are the corrective signal. The challenge is the enormous delay — rate changes take 12-18 months to propagate through the economy.
Key Insight
Monetary policy is a feedback loop with an 18-month delay — which is why it oscillates. Long delays make any feedback system prone to overshoot and oscillation, whether it's a shower faucet or an economy.
Sprint Retrospectives
Agile Software DevelopmentAgile methodologies build feedback loops into the development process. Sprint retrospectives measure team performance, compare it to goals, and generate process adjustments. However, the loop is often weak — retrospectives happen infrequently, measurements are subjective, and corrective actions are not always followed through.
Key Insight
Agile's genius was recognizing that software development is a control problem — but most teams run their feedback loops too slowly and with too much noise to achieve real stability.
Adaptive Signal Timing
Urban TrafficMost traffic signals still operate on fixed timers — green for 30 seconds, red for 45 — regardless of actual traffic conditions. The sensors exist (induction loops, cameras, connected vehicles), but the feedback loop is rarely closed in real time. Adaptive signal control systems that adjust light timing based on live traffic flow can reduce congestion by 10-25%, but fewer than 5% of intersections in the US use them.
Key Insight
A traffic light on a fixed timer is an open loop — it's the equivalent of a thermostat set to "always on." The data to close the loop already exists; the infrastructure just hasn't caught up.
Personalized Learning Loops
EducationTraditional classrooms deliver content at a fixed pace — the educational equivalent of a fixed-timer thermostat. Some students are bored, others are lost, and the teacher has limited ability to sense where each student is. Adaptive learning platforms that continuously assess mastery and adjust difficulty in real time are beginning to close this loop, but most education still runs open-loop.
Key Insight
A lecture is a broadcast, not a feedback loop. Education that doesn't continuously measure and adapt to the learner is flying blind — and wondering why it crashes.
Precision Irrigation
AgricultureMost farms irrigate on a schedule rather than in response to soil conditions. Precision agriculture closes the loop: soil moisture sensors measure water content, a controller compares it to optimal levels for the crop stage, and smart valves adjust water delivery in real time. Early adopters report 20-40% water savings with higher yields — the same efficiency gains that PID controllers brought to manufacturing a century ago.
Key Insight
Agriculture is the last major industry still running mostly open-loop. The transition from schedule-based to sensor-based farming is the same revolution that happened in manufacturing when feedback control replaced manual adjustment.
Related Patterns
Backpressure is a specific type of feedback loop where the signal flows upstream from consumer to producer, regulating flow rate rather than a general setpoint.
Feedback loops provide the sensing and response mechanisms that detect failure thresholds and trigger containment — without feedback, you can't know a compartment has been breached.
Rate limiting is often implemented as a feedback mechanism — the system measures current throughput and adjusts acceptance of new requests based on the gap between actual and maximum rates.
Active probing is a feedback loop specialized for unknown environments — send a signal, measure the response, update your model. Echolocation, sonar, and A/B testing are all feedback loops where the system generates its own stimulus rather than passively observing.
Arms races are feedback loops between adversaries — each adaptation by one party is sensed and countered by the other. The loop never converges to a setpoint because the "desired state" keeps moving. Every lock breeds a better lockpick breeds a better lock.