Poolside – Weekly Recap

Poolside advanced its enterprise AI strategy this week, centering on the rollout of its Laguna family of agentic coding models and the launch of a new enterprise AI agent platform. The developments underscore a push to move customers from experimentation toward production-grade deployments while maintaining strict security and governance.

The Laguna lineup is led by Laguna M.1, a 225B-parameter Mixture-of-Experts model with 23B activated parameters, and Laguna XS.2, a 33B-parameter MoE model with 3B activated parameters released under the Apache 2.0 license. Poolside positions Laguna XS.2 as its first open-weight model and highlights open weights as core to its enterprise strategy rather than an adjunct.

To broaden access, Poolside is distributing Laguna via its own API, an agent harness called pool, and a browser-based environment named Shimmer, while also partnering with Hugging Face, OpenRouter, Baseten, and Ollama. These channels are intended to reduce friction for developers and accelerate feedback loops that can refine long-horizon coding capabilities.

Performance and time-to-market metrics were another focus, with Poolside highlighting latency figures for Laguna XS.2 and Laguna M.1 delivered through infrastructure partner Baseten. Median time-to-first-token was cited at 146 milliseconds for Laguna XS.2 and 605 milliseconds for Laguna M.1, with P90 latency of 1.5 seconds and 3.9 seconds, respectively, under production-like conditions.

Poolside noted that it moved from project kickoff to a production-grade, white-labeled API in seven weeks using Baseten’s newly launched Frontier Gateway infrastructure. Developers access a Poolside-branded endpoint while Baseten manages the underlying stack, allowing Poolside to concentrate on model development and product differentiation while leveraging specialized inference infrastructure.

In parallel, the company introduced the Poolside Platform, a production-grade system for running AI agents inside an organization’s own security boundary. Customers can deploy models on bare metal, on air-gapped hardware via partners such as Dell Technologies, or within existing VPC setups on AWS, Azure, or Google Cloud, targeting regulated or security-sensitive sectors.

The platform emphasizes governance and control features, including containerized execution, built-in secret management, network policy controls, and full trajectory logging of agent actions and reasoning steps. Poolside’s agents integrate with tools such as VS Code, Visual Studio, IntelliJ, Slack, Jira, GitHub, Workday, and Salesforce via centrally managed MCP servers, embedding AI into existing workflows.

To support adoption, Poolside offers Forward Deployed Research Engineers who embed with customers to adapt the platform and deliver initial automated workflows within weeks. This high-touch model may accelerate time-to-value but could be more resource-intensive, requiring careful scaling to maintain margins as the customer base expands.

Collectively, the week’s announcements position Poolside as an API-first, enterprise-focused AI infrastructure provider with an emphasis on open-weight models, low-latency inference, and secure, controllable agent deployments. If enterprises validate the Laguna models and platform for mission-critical use cases, Poolside could strengthen its competitive position and expand recurring revenue opportunities in the emerging market for enterprise AI agents.