Act-On vs Gumloop

Gumloop has moved from a workflow automation platform to a multi-agent system with increasingly autonomous capabilities. In early September, the product introduced Gumball—a self-improving, proactive agent running on company infrastructure—background subagents that execute concurrently without blocking the main chat, and a Model Router that picks the right model per task automatically. These are not incremental additions; they represent a shift from 'automations you configure' to 'agents that improve themselves and manage their own execution resources.'

The self-improvement loop is the defining arc: Gumball learns from corrections, agents can now manage their own evaluation criteria, and the Model Router removes model selection as a user responsibility. The daily intelligence layer (Daily Chew, Gumball) is getting tighter with Highlights and actionable briefs. Gumloop is building toward autonomous contributors—proactive, self-assessing, and model-aware—rather than user-triggered task runners.

The next move is likely deeper agent-to-agent orchestration—routing work between Gumball and specialized subagents based on task type—or expanding the self-improvement loop to cover skill acquisition from past successful runs.