Cayu is a production-grade Python framework for building domain-specific AI agents with full lifecycle control, including context assembly, tool invocation, durable state management, and human oversight. Extracted from a software factory that deployed thousands of applications using specialized AI agents, Cayu addresses production challenges like process failures, state persistence, and cost attribution that existing agent SDKs overlook.
A programmer discovers that fsync() doesn't guarantee data durability the way commonly assumed, as multiple storage layers between application and disk have different definitions of "done". The article explains how write() only puts data in the page cache, not on disk, and how journaling filesystems add complexity to data persistence guarantees.
Oh My Subagents is a local runtime that enables durable, persistent delegation between parent and child AI agents for Codex and Claude users, maintaining state across interruptions and providing accountability through workflow definitions and immutable task assignments. It moves coordination into persistent runtime state rather than relying on polling loops or chat history, allowing managers to choose execution patterns dynamically while preserving complete checkpoints and audit trails.
A ranking of iPhone Pro materials—titanium, stainless steel, and aluminium—across five dimensions: feel, look, features, durability, and weight. Titanium excels in feel and durability; stainless steel in appearance; aluminium in features and thermal performance. The author reflects on how material choices affect the user experience across generations.
AWS Lambda now supports 90-minute function timeouts for asynchronous and event source mapping invocations on Lambda Managed Instances, a sixfold increase from the previous 15-minute limit. This enables longer-running workloads like media transcoding, financial calculations, and AI inference without requiring application re-architecture. Synchronous invocations retain the 15-minute maximum timeout.
Wiggle is a durable workflow engine that models business processes as graphs and executes them as state machines surviving crashes and worker failures. It features cellular architecture for multi-tenant isolation, pull-based gRPC workers supporting multiple languages, and can run embedded or distributed without external dependencies.