Tag
Using Pi Durable as a starting point, the article breaks down the evolution of the Agent Harness architecture: decoupling the persistence system, agent loop, and execution environment to enable lossless state recovery, codemode tool calls, and server/client separation. It also proposes a scalable architecture for a long-running Harness, with the goal of reducing hardware resource consumption by an order of magnitude.
作者将 1982 年 64KB PDP-11 Unix 的 init.d、fork/wait 与 overlay 机制类比到 LLM 编码代理的上下文压缩问题，通过编号状态文件（write-once 锁存、append-only 负记忆、WAL 式回放）让代理在 200k-token 压缩后不丢失关键不变量。
Armin Ronacher details how Pi Durable enables agent processes to resume from the point of interruption after a crash: persistence is achieved through an "effect sandwich" (intent, execution, result storage) rather than event sourcing, combined with a versioned document store to manage agent application state, supporting deployment to platforms such as Vercel, Cloudflare DO, and E2B.
The author open-sources mishe-tauftauf, a small local coordination system for coding agents based on 'observability-oriented programming,' where ephemeral agents ('minds') work from shared state documents, a text tape, and checks returning GREEN/RED/UNKNOWN instead of preserving agent continuity. The approach was validated in production on an adtech DSP built by a three-person team reaching ~50k RPS, and the project is released CC0 with a call for others to stress-test it in unfamiliar environments.
Google open sourced AX, an internal agent runtime built on Agent Substrate, which stores task state in Redis instead of Kubernetes/etcd to handle the churn of millions of short-lived agent tasks.
Jared Palmer announces Kev, a new open-source React state management library, introduced via a blog post with links to the project and source code.
A developer discussion of the tradeoffs between keeping tool-call working state in the LLM (simpler prompts but messy retries) versus owning it in the application (cleaner control flow but less agentic), highlighting duplicated or stale tool calls after timeouts as a key failure mode.
The article argues that OpenAI's Agents API does not make agent frameworks obsolete, citing reasons such as the need for model portability, state control, failure handling, and evaluation in production systems.
Release of VeriLoop E2, a 27B LLM post-trained from Qwen3.8-27B with VeriLoop-Governed Recurrence (VGR) for managing state transitions and generating training signals.
This paper introduces IntentFlux, an executable benchmark for measuring intent drift in multi-turn LLM agents, and StateForge, an explicit active-requirement maintenance method that partially improves task scores when users change their minds mid-dialogue.
The article explores common challenges in running multiple AI agents, such as debugging, state management, and cost, and invites community input on real-world experiences in production.
This article introduces the Chief of Staff Pattern, an organizational approach to orchestrating AI coding agents by separating coordination from execution to improve reliability and state management in long-horizon tasks.
ADK is a tool that gives developers fine-grained control over AI agent behavior by managing orchestration, tool use, and state in code rather than configuration.
The article argues that terminal sessions should not manage AI agent tasks and introduces Vincent, a daemon-based system that provides durable execution and state management for agentic workflows.
Infra is a provider-neutral continuity layer for AI agents that preserves authoritative state across interruptions, reducing model-token use by 49.4% and accelerating recovery by 60.8% in internal controlled tests.
The article explains how to manage complex application state in growing reactive UIs by using building blocks like Datastar, glimmer, Domino, and Ebb to handle data flows, business logic, and external integrations.
The article discusses the challenges in developing AI systems with persistent memory, emphasizing the need for effective state management beyond simple retrieval methods like RAG.
Solid Objects is an open-source library that implements a durable objects model for PostgreSQL, SQLite, and MySQL, providing stateful, ordered processing without vendor lock-in or additional infrastructure.
The article summarizes community insights on multi-agent orchestrators, highlighting the adoption of deterministic triggers, file-based state management, and cross-provider validation to improve workflow reliability.
The article discusses the challenges of testing AI agent workflows in sandbox environments versus production, highlighting issues like silent failures, state management, and the inadequacy of current testing methods, and seeks community advice on best practices.