AI agents are moving on-chain, and the platforms that serve them must scale elastically while staying secure for decades. NEAR’s sharded design and account model put it in a unique spot as builders rethink infrastructure for autonomous apps and data-heavy workloads. This article unpacks why two ideas—dynamic resharding and quantum-safe signing—are central to NEAR’s AI moment. You’ll learn how they work, what they change for developers, the risks to weigh, and pragmatic steps to prepare. Quick Answer NEAR’s dynamic resharding is about automatically splitting and merging state shards to match demand, which can smooth fees and throughput for spiky AI workloads. Quantum-safe signing refers to adopting post-quantum cryptography (PQC) or hybrid keys so accounts remain secure if powerful quantum computers emerge. Together, they point to an elastic and future-resilient stack: scale when agents surge; rotate keys as cryptography evolves. - Elastic capacity: shards adapt to load instead of forcing apps to migrate manually. - Better UX: steadier fees and lower congestion during AI-driven bursts. - Crypto agility: account-level key rotation enables gradual PQC adoption. - Risk-aware: resharding adds cross-shard design complexity; PQC adds overhead. How does NEAR’s dynamic resharding actually work? NEAR’s core architecture, Nightshade, partitions state across multiple parallel shards, with validators producing “chunks” for each shard and assembling them into blocks. Dynamic resharding is the capability to automatically split a hot shard into multiple shards or merge underused shards, based on network conditions. The intent is to keep capacity responsive without manual coordination by app teams. In practice, this kind of elasticity aims to reduce hotspots. When an AI-driven app spikes (think inference payments or agent-to-agent bidding), the shard hosting its state can be split to distribute load, while quieter periods allow merging to reduce overhead. This helps smooth fees and confirms transactions faster under pressure. NEAR has