The Age of Async Agents — Cognition's Walden Yan & OpenInspect's Cole Murray

2026-05-27 · Source: Latent.Space - Www.latent.space · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Cloud Computing & IT Infrastructure · Depth: Advanced, extended

Summary

The engineering world is rapidly shifting towards autonomous background agents, driven by advanced models like Opus 4.5 and GPT 5.2, which achieved significant capabilities around December 2025. This shift enables agents to autonomously drive tasks from specification to pull request with minimal human intervention, as demonstrated by Cognition's Devin, which saw a 7x increase in merged PRs and an 80% commit percentage in March from 16% in January. OpenInspect, an open-source project, emerged to address friction points in cloud-based agent sessions, particularly regarding context sharing. Key architectural decisions involve running the agent "in the box" versus "out of the box," with the latter, separating the agent's "brain" from the machine, preferred for security and infrastructure reuse despite its complexity. Managing the agent's working environment, including dependencies and credentials, remains a critical challenge.

Key takeaway

For AI Architects and MLOps Engineers designing or deploying autonomous coding agents, prioritize an "out-of-box" architecture that separates the agent's intelligence from its execution environment. This enhances security and allows for greater reuse of existing dev infrastructure. Focus on robust environment management and integrate agents deeply into your company's ecosystem, including read-only access to production systems and knowledge bases, to maximize their value for SRE, non-engineer contributions, and customer support.

Key insights

Autonomous background agents, powered by advanced models, demand robust infrastructure and careful architectural choices for security and scalability.

Principles

• Separate agent "brain" from machine for enhanced security.
• Optimize VM boot times using diff-based file systems.
• Local developer environments simplify agent setup and testing.

Method

The preferred agent architecture separates the agent's "brain" (control plane) from the sandbox environment, allowing the sandbox to act as "hands" for tool calls. This requires managing state externally but improves security and reusability.

In practice

• Deploy agents for SRE auto-triage and first-response scenarios.
• Enable non-technical roles (PMs, marketing) to initiate code changes.
• Integrate agents with production logs and databases for full visibility.

Topics

• Async Agents
• Cloud Agents
• Devin
• OpenInspect
• Agent Architecture
• VM Management
• SRE Automation

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, MLOps Engineer, AI Architect

Related on AIssential

• What we’ve learned building cloud agents — Cloud agent effectiveness hinges on robust development environments and durable, decoupled execution, shifting from local extensions to an operating layer.
• The Month AI Woke Up — AI agents have matured to enable autonomous task execution, fundamentally altering programming and impacting market valuations.
• 10 OpenClaw Lessons for Building Agent Teams — Agentic AI systems like OpenClaw offer significant utility but require deliberate design for task separation, security, and cost management.
• Agent Swarms and Knowledge Graphs for Autonomous Software Development with Siddhant Pardeshi - #763 — Autonomous development systems can deliver production-ready software at enterprise scale by orchestrating AI agent swarms.
• The Race to Put AI Agents Everywhere — AI agents are now viable, driving a rapid "clawification" of software and a sprint to enterprise productization.
• Build AI Coding Agents that Run ∞ in the Cloud — New tools empower AI coding agents with enhanced web access, workflow adaptability, and cloud-native execution.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Latent.Space - Www.latent.space.