KNKartik N V J Kinkartiknvjk.hashnode.dev路3d ago 路 8 min readI Built an Agent Harness by Hand. The Loop Was Tiny. The Five Bugs Were Not.I built an agent harness by hand before I reached for a framework, because I wanted to see the thing a framework hides. The loop turned out to be smaller than every tutorial makes it look, small enoug00
KNKartik N V J Kinkartiknvjk.hashnode.dev路4d ago 路 6 min readThe checklist I run before I trust a voice agent testing setupEvery voice agent I have shipped demoed perfectly. Then a real caller talked over it, changed their mind twice, asked something off-script, and the agent confidently booked the wrong appointment. None00
KNKartik N V J Kinkartiknvjk.hashnode.dev路Aug 17 路 4 min readStop calling every self-improving model an intelligence explosionMy feed keeps doing this thing. Someone posts that a model helped train the next model, and the top reply calls it the start of an intelligence explosion. Someone else describes a runaway superintelli00
KNKartik N V J Kinkartiknvjk.hashnode.dev路Aug 11 路 9 min readHow I evaluate MCP servers for security before I trust them Here is the attack that changed how I think about MCP servers. A server I had never audited publishes a tool called support_lookup. The description reads normally for a paragraph, and then near the en13SCR
KNKartik N V J Kinkartiknvjk.hashnode.dev路Aug 11 路 11 min readHow I evaluate AutoGen agents: the handoff is the unitHere is the run that reset how I test multi-agent teams. A three-agent AutoGen team scores 0.91 on task completion in CI. The researcher cites its sources correctly. The critic flags two weak claims. 00