Neuromorphic photonics has emerged as a promising route for brain-inspired information processing, with optical neural networks offering a path beyond key bandwidth and energy bottlenecks of conventional electronic hardware. Within this landscape, photonic reservoir computing has attracted growing interest because it reduces training complexity while retaining the rich dynamics needed for machine-learning tasks. Recent progress has expanded the field through improved integration, better control of memory and nonlinearity, and new strategies for parallel information processing. Yet delay-based single-node reservoirs still face major limitations, since virtual nodes are generated sequentially and throughput remains tied to cavity latency. Here, we present and discuss our recent work on a tunable neuromorphic photonic reservoir based on a nonlinear amplifying loop mirror and place it within the broader context of recent advances in photonic reservoir computing. The device combines controllable nonlinearity and fading memory with concurrent wavelength processing and dense temporal encoding, allowing multiple data streams to be processed within the same physical cavity. We also review recent developments in the field and discuss current challenges, emerging opportunities, and future directions toward scalable, low-latency, and energy-efficient photonic machine-learning hardware.