composability

The author explains why effect systems in programming languages are exciting, offering better control over resource interactions, composability, and testability compared to current systems.

This paper introduces HPSE, a hybrid-policy self-editing method for composable unstructured knowledge editing in LLMs, enabling edited models to answer atomic questions and perform multi-hop reasoning about injected knowledge. The approach uses proactive self-distillation with hybrid rollouts and is validated across multiple LLM backbones and editors.

An analysis of how platform engineering must evolve for AI agents, shifting from rigid golden paths to composable, API-first building blocks that support non-human identities, scoped permissions, and audit trails.

Kent C. Dodds shares an article and YouTube video arguing that the quality of AI agent outputs depends on the design of system primitives—providing agents with well-crafted, composable building blocks narrows their action space and yields reliable results.

This article introduces Onyx, a VM for programmable agent orchestration, turning agent orchestration into software engineering by providing a runtime with clear semantics and execution model. It discusses the design decisions and limitations of current agent systems, and presents the PROGRAMS specification.

The article discusses how LLMs have grown increasingly complex, moving beyond simple transformer stacks to incorporate diverse attention variants, mixture-of-experts, and multimodal encoders, drawing parallels with recommendation systems and emphasizing the need for composable kernel optimization like FlexAttention.

Phinite launches as a multi-agent OS infrastructure layer providing first-class agent identity, composable skills, behavioral evaluation, and cloud-agnostic deployment with built-in observability.

A Twitter thread discussing whether a database filesystem abstraction (PostgresFS) or a skill-based approach with local Bash is better for agent workflows. The skill approach wins on composability and speed.

The tweet discusses the problem of bloat in AI agent harnesses, agreeing with Kaffu's critique that harnesses become "rich man's toys," and advocates for a composable architecture of small, replaceable workers to reduce drift and keep systems cheap and debuggable.

This paper presents a technique for composing data types and functions from independent components, and extends the approach to combine free monads, enabling a modular structuring of Haskell's IO monad.

Matt Pocock asks for feedback on his proposed principles for skills: concise, single-responsibility, composable, progressively disclosed, and harness-agnostic.

Caidan Williams announces an upcoming update to terminal graph, highlighting improved composability.