Sign InOpen Brain
AI EngineerVideoSource Linked

Always-on agents run production without the on-call tax — Justin Smith, Resolve AI

Always-on production agents can turn releases and operational events into tailored monitoring tasks. The useful pattern is persistent context plus scoped triggers, checks, and escalation.

AI Engineer · Aug 9, 2026
Open Source Open MarkdownOpen JSON
Source Summary

Resolve describes cloud agents that run on schedules, event streams, or direct messages inside a sandbox with shared memory. In the demo, a **GitHub release tag** triggers inspection of the code changes, selection of relevant telemetry, and a release-specific monitoring plan.

Practical Implication

Start with bounded background work such as weekly handovers, health summaries, release watching, or answering operational questions. Connect existing skills and MCP services, and require human confirmation when the agent lacks enough confidence to respond.

Agent-Ready Context
Resolve describes cloud agents that run on schedules, event streams, or direct messages inside a sandbox with shared memory. In the demo, a **GitHub release tag** triggers inspection of the code changes, selection of relevant telemetry, and a release-specific monitoring plan.

Start with bounded background work such as weekly handovers, health summaries, release watching, or answering operational questions. Connect existing skills and MCP services, and require human confirmation when the agent lacks enough confidence to respond.

The release example used a **fake environment**, and the talk reports no measured reliability, cost, or incident-reduction results. Persistent access to telemetry, Slack, and production knowledge also makes permission scope, stale memory, and false confidence important unresolved risks.
Connected Context · Feed7 Judgment

This extends cloud agents from requested jobs to persistent operational participants triggered by schedules and events. It narrows the sensible adoption path to bounded monitoring and handover tasks with confidence-based escalation, while making memory quality, least-privilege access, traceability, and cost limits prerequisites for unattended operation. Because the demonstration was synthetic, production reliability remains unconfirmed.

Scaling Managed Agents: Decoupling the brain from the handsAnthropic’s separation of harness and replaceable sandboxes provides an infrastructure pattern for running the persistent background work described here.MemLens: A Value-Aware Memory Management System with Interactive Analytics for LLM-based AgentsMemLens directly addresses the stale or indiscriminate shared-memory risk by making individual memories and their tradeoffs inspectable.Agent Runs now available in the Vercel MCP and CLIRun traces covering tool calls and token use provide the observability needed to investigate and refine unattended operational agents from real behavior.AI Gateway now supports team and project spend budgetsHierarchical spend caps add a necessary failure boundary for scheduled and event-driven agents that can otherwise consume resources without an active operator.
Context Map
infracoding#observability#cloud-agents#agent-memory
Uncertainty
The release example used a **fake environment**, and the talk reports no measured reliability, cost, or incident-reduction results. Persistent access to telemetry, Slack, and production knowledge also makes permission scope, stale memory, and false confidence important unresolved risks.