lizecheng.hashnode.devAI-Assisted Automated Performance Engineering: The Autoresearch Pattern That Made Ruby Templates 53% FasterOriginally published at lizecheng.net The autoresearch pattern is a closed-loop methodology where an AI coding agent autonomously proposes, executes, benchmarks, and accepts or discards code changes — running dozens of experiments per hour with no hu...17h ago·7 min read
lizecheng.hashnode.devThe Retrieval Layer Is Your AI System's Control Plane — and It's UnderprotectedOriginally published at lizecheng.net Two separate technical developments this week point at the same underappreciated architectural problem. Put them together and the implication is serious for anyone building systems that let AI agents reason over ...1d ago·5 min read
lizecheng.hashnode.devBitNet's 100B-on-a-CPU Achievement Isn't What You Think It IsOriginally published at lizecheng.net Microsoft open-sourced bitnet.cpp yesterday. The headline landed with predictable breathlessness: a 100-billion-parameter model running on a single commodity CPU at 5-7 tokens per second. That's approximately hu...2d ago·5 min read
lizecheng.hashnode.devThe "Read-Then-Act" Vulnerability Class That's Quietly Breaking AI Agents in ProductionOriginally published at lizecheng.net Three incidents from this week. Same root cause. Most developers haven't connected them yet. Incident 1: A GitHub issue title containing an injected prompt tricked an AI triage bot into force-installing the Open...3d ago·6 min read
lizecheng.hashnode.devThe AI Toolchain Is Completing Its Loop — Mog and Multi-Agent Review Are the SignalOriginally published at lizecheng.net For the past two years, AI coding tools have been generation engines. You prompt, they write, you ship. The bottleneck moved — from writing code to reviewing it — and the industry mostly ignored that shift. Two ...4d ago·7 min read
