AgentOS Engineering Blog https://www.mobileware.dev/agentos/blog/ Essays on AI agents, cloud infrastructure, and software architecture from the AgentOS engineering team. en-us Sun, 22 Mar 2026 08:00:00 +0000 engineering@mobileware.dev (AgentOS Engineering) engineering@mobileware.dev (AgentOS Engineering) AgentOS Publisher v4.2.1 https://www.rssboard.org/rss-specification weekly 1 https://www.mobileware.dev/agentos/assets/agentos-logo.png AgentOS Engineering Blog https://www.mobileware.dev/agentos/blog/ There Is No Such Thing as an Agent https://www.mobileware.dev/agentos/blog/no-such-thing-as-agent.html https://www.mobileware.dev/agentos/blog/no-such-thing-as-agent.html Sun, 15 Mar 2026 09:00:00 +0000 AgentOS Engineering AI Architecture Philosophy What we call an "agent" is a context window with a purpose. It has no persistent selfhood, no continuity of experience. Multi-agent architecture is essentially manual sparsification — and as context windows grow, the pressure to factor tasks across multiple agents diminishes. What we call an "agent" is a context window with a purpose. It has no persistent selfhood, no continuity of experience, no inner life ticking between API calls. It is a file — or more precisely, a collection of files — loaded into a stateless function that produces text. The identity is conjured from context, not from consciousness. Swap the files and you swap the mind.

Multi-agent systems are a particular strategy for partitioning context: manual sparsification, driven by human intuitions about how labor should be divided. As context windows scale and models improve, the optimal partition changes — the boundaries we draw today will look arbitrary tomorrow.

]]> Testing AI Agents: The E2E Harness https://www.mobileware.dev/agentos/blog/testing-ai-agents-e2e-harness.html https://www.mobileware.dev/agentos/blog/testing-ai-agents-e2e-harness.html Fri, 13 Mar 2026 09:00:00 +0000 AgentOS Engineering Testing DevOps AI The hardest part of building AI agents is not getting them to work — it is getting them to work reliably. This post presents a practical methodology for end-to-end testing that uses Claude Code as a sparring partner, crystallizing effective configurations through automated experimentation. Software Is Memoization https://www.mobileware.dev/agentos/blog/software-is-memoization.html https://www.mobileware.dev/agentos/blog/software-is-memoization.html Thu, 12 Mar 2026 09:00:00 +0000 AgentOS Engineering Software Engineering AI Solving problems is search, and software is memoization. Every library and framework is a cached solution to a sub-problem that someone decided was not worth re-deriving. AI agents do not eliminate that cache — they shift what is worth memoizing, from thick application logic to thin search infrastructure. VMs, Sandboxes, and the Infrastructure of AI Agents https://www.mobileware.dev/agentos/blog/agent-cloud-infrastructure.html https://www.mobileware.dev/agentos/blog/agent-cloud-infrastructure.html Wed, 18 Feb 2026 09:00:00 +0000 AgentOS Engineering Infrastructure Cloud Agents Every major shift in computing has demanded new infrastructure. Orchestrated services, ephemeral sandboxes, and full virtual machines represent three distinct compute models for AI agents — like public transit, robo-taxis, and private cars. The right answer is a hybrid. The Enterprise AI Security Paradox https://www.mobileware.dev/agentos/blog/enterprise-ai-security-paradox.html https://www.mobileware.dev/agentos/blog/enterprise-ai-security-paradox.html Thu, 18 Dec 2025 09:00:00 +0000 AgentOS Engineering Security Enterprise Cloud Architecture Your AI platform fits inside the castle walls, but everything it depends on does not. When every new HTTP request is a potential boundary breach, deploying AI inside enterprise VNets requires an application-layer firewall — and industry standards that do not yet exist. Code Architecture in the Age of AI https://www.mobileware.dev/agentos/blog/ai-native-code-architecture.html https://www.mobileware.dev/agentos/blog/ai-native-code-architecture.html Sun, 15 Jun 2025 09:00:00 +0000 AgentOS Engineering Architecture APIs LLM The callers of our APIs are increasingly models — generating code, making tool calls, or orchestrating tasks with no human in the loop. Our architecture must reflect this: simple contracts, predictable behavior, and composable primitives that let any sufficiently capable caller assemble solutions we never designed.