AI Agents on Bruce on AI Engineering

Sub-Agent Architecture for AI Coding Harnesses: When to Spawn, How to Route, What It Costs

Mon, 13 Apr 2026 10:00:00 +0800

This is Part 3 of the Harness Engineering series. Part 1 framed the thesis (Agent = Model + Harness). Part 2 went deep on CLAUDE.md, the single most important feedforward control. This article goes deep on a structural decision most teams get wrong: when to spawn a sub-agent, how to route it, and what it actually costs.

Harness Engineering #2: The 60-Line CLAUDE.md Rule (and Why My 90-Line File Failed)

Tue, 31 Mar 2026 10:00:00 +0800

This is Part 2 of the Harness Engineering series. Part 1 framed the core identity: Agent = Model + Harness. Part 3 goes deep on sub-agent architecture. This piece sits between them, on the single highest-ROI file you will ever write for a coding agent — and the one most teams write wrong.

I have read a lot of CLAUDE.md files on GitHub over the past three months. Most of them fail the same three ways: they are too long, they were generated by an LLM, and they live in a single flat file when the project has outgrown that shape. I know all three failure modes intimately because I shipped all three in my own repo. This guide is what I wish someone had handed me before I spent a weekend debugging why my agent had gotten dumber.

Harness Engineering: After 60 Days Running an AI Coding Pipeline, the Model Was the Least Important Part

Mon, 30 Mar 2026 10:00:00 +0800

This is Part 1 of the Harness Engineering series. Part 2 goes deep on CLAUDE.md best practices. Part 3 covers sub-agent architecture and model routing.

I have been running an AI coding harness in production for 60 days — the pipeline that writes, illustrates, and distributes posts for this blog. The single most surprising finding from that experiment: the model is the least important part of the agent. Not irrelevant, but far from dominant. Upgrading the writer from Sonnet to Opus lifted blind-rated output quality by about 5%. Rewiring the harness — routing, context boundaries, sensors — cut end-to-end cost by roughly 60% while quality went up, not down.

OpenAI Symphony: From Issue Ticket to Pull Request Without a Developer

Fri, 06 Mar 2026 22:00:00 +0800

What if your issue tracker could fix its own tickets?

OpenAI just open-sourced Symphony, an automation service that monitors your project’s issue tracker (like Linear), spawns autonomous coding agents (like Codex) for each task, and delivers verified pull requests — complete with CI status, code reviews, and walkthrough videos — before a human even looks at the code.

This isn’t another AI code assistant that waits for your prompt. Symphony represents a fundamental shift in how software gets built: from developer-driven coding to project-driven orchestration. In this guide, we’ll break down Symphony’s architecture, explain how it works under the hood, and show you how to set it up for your own projects.

OpenClaw vs CrewAI vs AutoGen 2026: 6 AI Agent Frameworks Ranked

Fri, 06 Mar 2026 19:00:00 +0800

The AI agent landscape in 2026 is crowded. OpenClaw exploded to 247,000+ GitHub stars in weeks. AutoGPT pioneered the autonomous agent concept. CrewAI simplified multi-agent Python workflows. LangGraph brought graph-based orchestration. AutoGen introduced multi-agent conversations. Devin showed what a fully commercial AI engineer looks like.

So which one should you actually use?

This comparison breaks down six major AI agent tools across architecture, setup complexity, multi-agent support, messaging integration, cost, and ecosystem. Whether you want a personal AI assistant on Telegram, a coding agent for your development workflow, or a multi-agent framework for your Python project, you will find a clear recommendation here.

OpenClaw Tavily Search: ClawHub Skill Integration Guide (2026)

Fri, 06 Mar 2026 14:00:00 +0800

Your OpenClaw agent is smart, but it is blind. Without web search, it generates answers from training data alone — which means confident-sounding responses built on outdated or fabricated information. Ask it about yesterday’s news and it will either admit ignorance or, worse, invent something plausible.

Tavily fixes this. It is an AI-optimized search API that returns clean, structured results designed specifically for AI agents — not the noisy HTML pages a human would browse. Integrating Tavily into OpenClaw gives your agent real-time access to the web, turning it from a knowledge-limited chatbot into a research-capable assistant.

OpenClaw Pitfalls: 15 Automation Mistakes and Fixes

Fri, 06 Mar 2026 09:00:00 +0800

You installed OpenClaw. You added Tavily search, proactive-agent, and a handful of community skills. Everything looked amazing for the first 48 hours.

Then things started breaking. Your agent hallucinated an entire research report because the search API key expired silently. Your monthly API bill hit $200 because Opus was handling heartbeat checks. A skill you installed from ClawHub had unrestricted shell access and started modifying files outside your workspace.

OpenClaw Setup Guide: Install and Configure Your AI Agent

Thu, 05 Mar 2026 09:00:00 +0800

Want a personal AI agent that runs 24/7 on your computer, responds through Telegram or WhatsApp, and actually executes tasks — not just chats? That is exactly what OpenClaw does. It is an open-source AI agent gateway that turns your machine into a command center for AI-powered automation.

This guide walks you through the complete OpenClaw setup process: from installation to connecting messaging platforms, configuring AI models, installing Skills, and locking down security. By the end, you will have a working personal AI agent you can message from anywhere.

OpenClaw Multi-Agent Setup: Build AI Teams That Work

Wed, 04 Mar 2026 19:00:00 +0800

You built an OpenClaw personal assistant. It handles your Telegram messages, searches the web, writes drafts, reviews code. Life is good.

Then reality hits. You ask it to research a competitor, and it responds with coding suggestions from yesterday’s conversation. You ask it to write an article, and it pulls in debugging context from a completely unrelated project. Your single agent is drowning in accumulated memory, and every response gets slower and more confused.

OpenClaw 2026.3.1: WebSocket Streaming, Agent Routing, and K8s Support

Tue, 03 Mar 2026 10:00:00 +0800

OpenClaw just dropped version 2026.3.1 — and this one matters. WebSocket-first streaming for OpenAI models, native Kubernetes health checks, a proper agent routing CLI, and Claude 4.6 adaptive thinking enabled by default.

If you’re running OpenClaw in production or considering it for your AI agent workflow, this guide covers every meaningful change in the release, what it means for your setup, and how to configure the new features.