Block introduces an on-device, open-source AI agent called Goose, which allows developers to choose their preferred LLM to automate engineering tasks
Context & Ripple Effects
Block's Goose brings the agent model directly into engineering work: it runs on-device, is open source, and leaves model selection to the developer rather than tying the workflow to one LLM provider.
It fits a growing set of locally run, multi-model agents, including Clawdbot's local, multi-LLM approach. The related coverage also shows the other end of the market: Google's enterprise platform for managing agent fleets targets centralized deployment and governance.
First-order effects
- Developers using Goose can automate engineering tasks while choosing the LLM that best fits their requirements, rather than adopting a fixed-model agent.
- Block makes the agent's on-device and open-source design available for inspection and adaptation, shifting more implementation control to developer teams.
Second-order effects
- LLM vendors must compete for use inside the same agent workflow on model quality, cost, and task fit, since Goose does not reserve the interface for a single provider.
- Teams evaluating coding agents gain a distinct trade-off between locally controlled, adaptable tooling and centrally managed enterprise-agent platforms.
Third-order effects
- If multi-model local agents gain adoption, the agent layer may become more separable from the underlying model layer, reducing the value of model lock-in for developer workflows.
- The market could split between open, developer-controlled agents and managed agent fleets, with governance and integration—not just model capability—becoming a key differentiator.
The trend: Developer AI is moving toward agentic work surfaces that can execute tasks across tools while preserving flexibility over the model underneath.