Author

Architect

Professional architect sharing high‑quality architecture insights. Topics include high‑availability, high‑performance, high‑stability architectures, big data, machine learning, Java, system and distributed architecture, AI, and practical large‑scale architecture case studies. Open to ideas‑driven architects who enjoy sharing and learning.

1.6k

Articles

Likes

10.0k

Views

Comments

Latest from Architect

100 recent articles max

Architect

Sep 30, 2026 · Artificial Intelligence

Jev: Separating Semantic Judgment from Generation in LLM Applications

TypeSafe's Jev introduces a decision model that returns typed probability outputs (Noul, Choice, Score) instead of natural language, moving semantic judgment from runtime generation to a designable, testable 'decision surface' developed with LLM assistance but executed by a specialized model, enabling calibrated thresholds, parallel evaluation, and deterministic code control flow.

Agent EngineeringDecision SurfaceJev

0 likes · 26 min read

Jev: Separating Semantic Judgment from Generation in LLM Applications

Architect

Sep 28, 2026 · Artificial Intelligence

Dream-RSI: How Agents 'Dream' on Past Searches to Optimize Exploration Budgets

Dream-RSI introduces a recursive self-improvement framework where AI agents improve exploration strategies by replaying historical search trees offline, enabling budget allocation decisions without retraining models, demonstrated across algorithm engineering, math optimization, and GPU kernel tasks with reduced search costs.

AI researchDream-RSIGoogle DeepMind

0 likes · 26 min read

Dream-RSI: How Agents 'Dream' on Past Searches to Optimize Exploration Budgets

Architect

Sep 27, 2026 · Artificial Intelligence

Plan Mode Evolved: The Verifiable Work Loop AI Agents Actually Need

The article argues that Plan Mode isn't obsolete but should shift from static pre‑approval documents to a dynamic, evidence‑updated work loop where agents explore, act, verify, and escalate high‑impact decisions to humans, illustrated with a sync‑to‑async export refactoring example.

AI AgentsClaude CodeCodex

0 likes · 39 min read

Plan Mode Evolved: The Verifiable Work Loop AI Agents Actually Need

Architect

Sep 26, 2026 · Artificial Intelligence

Meta Muse Architecture: Inside the Persistent Personal Agent Execution Environment

This article analyzes Meta's Muse architecture, a persistent personal AI agent execution environment running on dedicated cloud VMs, detailing its Runtime Cell isolation, Sentinel authorization control plane, credential proxy injection, tainted egress protection, browser isolation, and multi-layer state management for long-running tasks, while comparing with Codex, OpenClaw, Hermes, and DSH approaches.

AI Agent ArchitectureCredential ProxyLong-running Tasks

0 likes · 36 min read

Architect

Sep 25, 2026 · Artificial Intelligence

From LLM Call to Full Harness: Understanding Agent Runtime Layers

The article traces the evolution from a simple LLM call to a complete Harness for AI agents, using a timeout-fixing task to illustrate the layered boundaries between model, loop, and harness, covering context engineering, tool execution, state management, verification loops, and design trade-offs in systems like Pi, Cursor, and Codex.

LLM RuntimeState ManagementTool Calling

0 likes · 34 min read

From LLM Call to Full Harness: Understanding Agent Runtime Layers

Architect

Sep 24, 2026 · Artificial Intelligence

Laya vs Nimble vs Kev: Choosing the Right Open-Source Jev for Agents

This article compares three open-source Jev implementations — Laya, Nimble, and Kev — analyzing their model architectures, training approaches, performance benchmarks, deployment requirements, and the four runtime components (context scoping, logging, policy code, failure loops) needed before using semantic decision models in production Agent systems.

Agent ArchitectureJevKev

0 likes · 26 min read

Laya vs Nimble vs Kev: Choosing the Right Open-Source Jev for Agents

Architect

Sep 23, 2026 · Artificial Intelligence

Supervisor Agent Pattern: Dynamic Task Routing, Bottleneck Risks, and Control Patterns

This article analyzes the Supervisor pattern in multi-agent systems, explaining how central agents dynamically route tasks based on intermediate results, comparing 'agents as tools' vs. handoff control patterns, detailing context engineering practices, termination conditions, logging requirements, and why Supervisor architectures can become information bottlenecks despite their flexibility.

AI Agent ArchitectureMulti-Agent Systemsagent orchestration

0 likes · 23 min read

Supervisor Agent Pattern: Dynamic Task Routing, Bottleneck Risks, and Control Patterns

Architect

Sep 22, 2026 · Artificial Intelligence

Jev: Extracting Semantic Micro-Judgments from Agent Loops into Observable Decision Layers

TypeSafe's Jev isolates semantic micro-judgments — routing, risk scoring, boolean checks — from generative LLM calls using Choice, Score, and Noul primitives that return typed probabilities, enabling policy code to branch directly on observable, auditable signals instead of parsing free-text model outputs.

Agent ArchitectureChoice Score NoulJev

0 likes · 25 min read

Jev: Extracting Semantic Micro-Judgments from Agent Loops into Observable Decision Layers

Architect

Sep 20, 2026 · Artificial Intelligence

Workflow in Multi-Agent Systems: Code-Controlled Routing, Node Checkpoints, and Replayable Pipelines

This article explains the Workflow architecture for multi-agent systems where code controls routing over fixed paths while agents handle node-internal reasoning, detailing checkpoint placement at node boundaries for retry/pause/recovery, required node output fields, event logging, and a comparison of how LangGraph, CrewAI, Microsoft Agent Framework, OpenAI Agents SDK, and Claude Agent SDK express such workflows.

CrewAILangGraphMulti-Agent Systems

0 likes · 20 min read

Workflow in Multi-Agent Systems: Code-Controlled Routing, Node Checkpoints, and Replayable Pipelines

Architect

Sep 19, 2026 · Artificial Intelligence

Claude Writes 80% of Anthropic's Code—Why That's Not True RSI

The article analyzes Anthropic's claim that Claude writes 80% of merged code, explaining that code volume doesn't equal recursive self-improvement (RSI). It breaks down RSI runtime boundaries—verification, memory, rollback—and examines RSIAgent and GLM case studies to show current systems lack autonomous research judgment and reusable, verifiable experience transfer.

AI AgentsAnthropicClaude

0 likes · 29 min read

Claude Writes 80% of Anthropic's Code—Why That's Not True RSI