From model weights to the worksite: AI's new working surfaces
An open-weight model, a collaboration app, a story tool and a construction robot reveal distinct layers of AI deployment.
Artificial Intelligence··Morning
The layer that begins with open weights
Poolside presents Laguna S 2.1 as a 118-billion-parameter coding model that activates roughly 8 billion parameters per token through a mixture-of-experts architecture. The company says it supports a context window of up to one million tokens, published the weights on Hugging Face under the OpenMDW-1.1 license, and completed training in under nine weeks. A no-login chat interface, hosted services and integrations with several coding tools provide multiple routes to the same model.[1]
Poolside supplied the Terminal-Bench 2.1 and SWE-Bench Multilingual results, and those figures have not yet been independently verified. That limitation separates open weights from verified performance. The distribution pattern in the record is still clear: the model is not confined to one company interface, because weights, hosted access and tool integrations are offered together. The technical layer available to downstream products is consequently broader than a single chat screen.[1]
Team conversation and personal stories
Jack Dorsey's Buzz was released as an open-source, model-agnostic desktop tool that puts human teammates and AI agents in the same conversation. Available free on macOS, Windows and Linux, it is designed to let teams manage GitHub projects from inside the chat. Dorsey positions it as a decentralized, self-sovereign alternative to Slack and GitHub. Its product claim is therefore centered less on a new foundation model than on a collaborative surface that brings existing models into team workflows.[2]
Meta's limited StoryKit pilot opens a different user context. Parents can turn a photo of a toy or person into a character, describe the story world, and incorporate values such as kindness, courage or empathy; the app also generates customizable music. Meta says there are no social features, AI safety filters are included, and account access is restricted to adults. Buzz emphasizes collaborative production, while StoryKit emphasizes personalization under adult control.[3]
Moving from screens to physical work
Gritt uses AI-controlled robots to unload, transport and place large solar panels with sub-millimeter accuracy. The company announced $32 million in total funding, including a $26 million Series A. According to Gritt's own figures, an eight-person crew using its systems can install 3,000 to 4,000 panels a day, compared with 800 without automation. That comparison comes from the company; the selected record does not provide an independent field validation.[4]
The company says it plans to expand into fastening panels, drilling posts, building racks and tying rebar. Placed beside the other three products, that plan shows AI deployment occurring at separate layers, from model weights to application interfaces and physical execution. None of the four records supplies a common adoption measure: pilot scope, open-source availability, vendor benchmarks and funding totals describe different things. The comparison therefore establishes the products and their deployment directions without claiming broad uptake or success.[1], [2], [3], [4]
