Archives
All the articles I've archived.
-
The Evolution of AI Debate: From Rigid Workflows to Autonomous Deep Agents
A retrospective on building our Debate System: from the rigid scripts of May 2025 to the autonomous, tool-wielding Deep Agents of today. How we achieved 'A2A' collaboration without a protocol.
-
Solving the Context Window Problem: Resilient Context Management in Long-Running Agents
Long-running agents inevitably hit token limits. Here's how we built a 'Self-Healing' middleware to catch overflows, summarize history, and replay requests automatically.
-
From Rigid Workflows to Deep Agents: A Paradigm Shift in Context Engineering
Why we abandoned our complex 'Workflow Graph' in favor of the 'Agent Loop'. A deep dive into how changing who manages context—Human vs. AI—unlocked true intelligence.
-
Teable + AI: Enhancing Agents with SQL and Automated Labeling
Why we integrated Teable (a hyper-fast Airtable alternative) into our Agent system. It's not just a database; it gives Agents SQL capabilities and a standardized pipeline for AI data labeling.
-
The Evolution of Agent Sandbox: From Docker to Cloud Services
Following the 'Manus Effect', code execution became a standard Agent capability. We share our journey from exploring self-hosted Docker/NFS to adopting cloud sandboxes (E2B, PPIO) and implementing a custom OSS filesystem backend.
-
My Journey with MCP: From RPC Wrapper to Native Agent Integration
A candid retrospective on adopting the Model Context Protocol (MCP). From early misunderstandings (using it like RPC) to understanding stdio vs. SSE, and solving the long-running task progress challenge.