Tag
Databricks introduces MemCurrent, a persistent memory architecture for production agents that powers Agent Bricks Memory and Genie, combining plugin, tool, and offline "dreaming" pipelines to triple pass rates while cutting tokens by 31% and latency by 26%.
Omar Santos presents a guide on using Jev, TypeSafe AI's decision model, as an LLM judge for agent evaluations, showing how reading full trajectories (tool calls, results, errors) rather than just final replies increases reliability and consistency for production-grade evaluation and monitoring systems.
An opinion post arguing that real production agent failures stem from uncontrolled execution of side effects rather than bad model outputs, proposing deterministic control points (approval gates, hard conditions, failure policies) before tool calls, and asking how the community handles this.
LangChain announces new Connection features in Managed Deep Agents 0.7, providing simple agent authentication with support for agent and user credentials to streamline deployment.
LangChain introduces Managed Deep Agents (MDA), a simplified way to build and deploy production AI agents with managed infrastructure and a deep agent harness.
Asks how developers budget agent retries to distinguish transient failures from persistent ones, and what signals best decide when to stop or retry in production agents.
A reflection on the challenge of managing permissions for AI agents in production, arguing that inventory alone is insufficient and that teams need unified control over agent actions, with ongoing interviews planned.
The article argues that production AI agents require a new platform abstraction beyond traditional PaaS, focusing on non-deterministic planning, durable execution, authorization, and auditability.
Built Trovis, a tool that records and explains agent actions in production, showing deviations from expected behavior.
The article explores loop engineering as a potentially missing discipline for developing robust production agents in AI systems.
The article argues that human-in-the-loop in agent systems should move from vague approvals to explicit, auditable step-level signed decision records with detailed evidence, payloads, idempotency keys, rollback paths, and ownership. It highlights the danger of approving a black-box story rather than a specific operation.
Eve is a framework for building durable, production-ready AI agents with built-in observability, introspection, and self-improvement capabilities, designed to work seamlessly with Next.js and Vercel deployments.
The article argues that diagnosis—explaining why an agent failed in operational terms and what is safe to do next—is a missing first-class skill in production agent stacks, more critical than making agents sound smart.
Morph Reflexes offers fast, cheap semantic signals from agent traces using multi-head classifiers on a shared backbone, enabling scalable monitoring of production agents.
The article discusses that the real challenge in AI agents is not building them but running them in production, and proposes the need for an operating system layer to manage fleets of agents, akin to how an OS manages software processes.
The article argues that production agent harnesses should not be monolithic frameworks but rather a stack of independent, replaceable workers connected by a shared trigger primitive, outlining 15 core responsibilities and how the iii engine implements this approach.
Google Devs introduces Agent Factory series with ADK 2.0 and Gemini 3.5 Flash, demonstrating how to build production-grade stateful agents that can run for days, featuring insights on skills, MCP, and code review strategies from engineers like Rohde Davis.
Anthropic introduces Claude Managed Agents, a set of composable APIs for building and deploying production-grade agents, addressing infrastructure challenges that separate prototypes from production.
A community discussion on agent memory reveals that while various patches exist for what to write down (e.g., plain files, layered memory, post-mortems), the unsolved problem is what to keep—detecting failures is tractable, but deciding which lessons persist still needs human judgment.
Experiments with a live agent processing market data through a governed runtime revealed three surprises: prompt structure drives execution reliability over reasoning quality; structured output can influence agent decisions; and separating reasoning and extraction into two calls maintains high parse success. The findings suggest governance belongs at the execution boundary, not on freeform reasoning.