NASA is testing a technology called Dynamic Targeting that would enable Earth-observing spacecraft to decide autonomously and within seconds where to make science observations from orbit. The concept shifts the decision-making loop from ground controllers — who plan observations hours or days in advance, uplink commands, and wait — to onboard AI that can react to what the satellite actually sees in real time. The operational constraint this addresses is fundamental. An Earth-observing satellite in low orbit crosses any given point in roughly seven minutes. If something scientifically interesting is happening below — a volcanic eruption, a sudden algal bloom, a wildfire igniting — the satellite either has pre-programmed instructions to observe it or it doesn't. By the time a ground team identifies the event, decides it's worth observing, and uplinks new commands, the satellite may have moved on. The observation window is gone. Dynamic Targeting collapses that decision loop from hours to seconds. The AI component isn't the flashy part — it's the enabling constraint. For a satellite to autonomously retarget, it needs to process sensor data onboard, evaluate what it's seeing against scientific priorities, decide whether to change its observation plan, and execute the manoeuvre — all with the limited compute and power budget of a spacecraft. This is edge AI in the most literal sense: compute at the physical edge of the network, where latency to the decision-maker (ground control) is the bottleneck being eliminated. The broader context is a shift in how NASA thinks about its observation architecture. Traditional Earth science missions are designed around fixed observation plans — orbits, swath widths, revisit rates. The satellite is a sensor platform; intelligence lives on the ground. Dynamic Targeting inverts this: the satellite becomes an agent with the ability to prioritise. This is the same architectural shift that autonomous vehicles represent for transportation — moving decision authority from a remote operator to the platform itself. The implications for Earth science are significant if the technology matures. Transient events — the phenomena that matter most for disaster response, climate monitoring, and ecological tracking — are by definition unpredictable. A constellation of satellites with Dynamic Targeting could function as a responsive observation network rather than a set of fixed cameras on predictable tracks. The science yield per orbit increases because the satellite can allocate its limited observation time to whatever is most valuable at that moment. The technology is in testing, not deployment. NASA's framing — "could enable" — is appropriately cautious. The gap between demonstrating autonomous targeting in a test environment and trusting it on a flagship mission with irreplaceable instruments is substantial. But the direction is clear: the era of satellites as passive platforms following ground-scripted plans is ending.
