Poolside Debuts Laguna AI Models and Opens New Chapter in Agentic Coding

According to a recent LinkedIn post from Poolside, the company is making its in-house AI model work public for the first time with the release of two Laguna family models, Laguna M.1 and Laguna XS.2. The post describes these as agentic coding models designed for long-horizon software tasks, such as writing code, running tests, inspecting failures, and creating tools.

The LinkedIn post highlights that Laguna M.1 is a 225B total parameter Mixture-of-Experts model with 23B activated parameters, trained from scratch on 30T tokens, positioned as Poolside’s most capable model to date. Laguna XS.2 is presented as a 33B total parameter MoE model with 3B activated parameters and is notable as an open-weight release under the Apache 2.0 license.

As shared in the post, Poolside emphasizes that these models emerge from an internal “model factory” built over several years, including proprietary data pipelines, training stack, and agent infrastructure. The content suggests deployment experience in demanding enterprise and government environments, which may help signal production readiness and credibility to potential commercial users.

The post also points investors to multiple access channels for Laguna, including the Poolside API, an agent harness called pool, and a browser-based development environment named Shimmer. Distribution partnerships with Hugging Face, OpenRouter, Baseten, and Ollama are mentioned, indicating a strategy to broaden developer reach and lower adoption friction.

From an investor perspective, the move toward public releases and open-weight licensing for Laguna XS.2 could accelerate ecosystem growth and community validation, potentially improving model quality and brand visibility. At the same time, the focus on agentic coding and long-horizon work may position Poolside in a higher-value enterprise software and developer-tools niche, where willingness to pay and switching costs can be significant.

The post’s description of contributions from roughly 60 people across applied research, infrastructure, and product suggests a relatively lean but specialized team executing a vertically integrated stack. If Poolside can convert this technical footprint and partner distribution into recurring enterprise and government contracts, the Laguna launch could mark an inflection point toward more predictable revenue and a stronger competitive stance in the AI model and tooling market.