AI digest: Infrastructure gets serious
Major infrastructure releases from IBM, LangChain, and Moonshot AI alongside enterprise governance tools and ByteDance's copyright troubles.
The tooling is maturing fast. This week brought serious infrastructure releases and enterprise-grade governance systems.
LangChain tackles multi-step agents with Deep Agents
LangChain released Deep Agents, a structured runtime designed for complex, stateful agent workflows. Most LLM agents fall apart when tasks become multi-step and artifact-heavy, which is exactly what Deep Agents aims to fix. This feels like the kind of infrastructure piece that could unlock more reliable agent deployments in production.
IBM ships compact speech model for edge deployment
IBM’s Granite 4.0 1B Speech targets multilingual speech recognition and translation for edge devices where memory and latency matter as much as accuracy. The 1B parameter size suggests they’re optimising for deployment constraints rather than chasing benchmark scores. Smart move given how many speech applications need to run locally.
OpenViking brings filesystem-style memory to AI agents
OpenViking from Volcengine organises agent context through a filesystem paradigm rather than flat text chunks. The idea is that agents need structured memory, resources, and skills management. Interesting approach, though we’ll see if the filesystem metaphor actually maps well to how agents think about context.
ByteDance shelves Seedance 2.0 over copyright concerns
Hollywood studios collectively forced ByteDance to pause the global launch of Seedance 2.0, their AI video generator. This shows just how convincing AI-generated video has become when it rattles the entire film industry. Legal pressure is becoming the real bottleneck for video AI deployment.