Platform teams are engineering‑led rather than product‑led, meaning there is typically no product manager providing a roadmap, no revenue target to chase, and no market pressure to lose. As a result, genuine work only exists when an engineer creates it themselves. The central challenge for a staff engineer is to decide what the team should build next by detecting signals that already exist in the environment.
Signals emerge from four domains. From the system, crash postmortems reveal what to fix or replace, and cloud‑bill cost analysis exposes buy‑versus‑build trade‑offs that should be revisited as technology and scope shift. From users, regular interviews uncover pains and existing hacks, while observing overloaded use‑cases shows how users repurpose the platform for problems it was never designed to solve, offering ready‑made prototypes worth assimilating when others share the same need. Partner‑to‑prototype lets engineers explore such solutions jointly with users.
From the organization, OKRs already encode agreed‑upon work, repeated mentions by managers hint at unaddressed concerns, and the debris left by laggard adopters after a migration points to gaps in the current offering. From industry, writing descriptive design documents of the existing system and comparing them with state‑of‑the‑art surfaces outdated decisions, and recognizing the cyclical bundling/unbundling of computing domains lets engineers import proven trends after early‑adopter risk has been taken elsewhere—a practice the author calls lag arbitrage. Because many signals compete, the author ranks them by how much argument they provide and whether they are leading or lagging; pre‑argued lagging signals are easy to act on but may be outdated, while leading signals that require constructing an argument offer fresh insight but need justification. The key skill is therefore not just spotting a signal but being able to explain why one signal merits attention over the others, turning the continuous struggle of inventing work into a disciplined practice.