research with AI
5 essaysResearch with AI #5: Fifty Agents, One Pair of Eyes
Usage limits start the story. Model families change what it's about.
"The obvious reason to run more than one AI model is the usage limit. The less obvious reason is what different model families notice — and what changes when they don't agree. Plus the practical bridges (Vox, codex-plugin-cc, kimi-plugin-cc, cc-fleet — and the oldest option of all) for consulting one model, delegating to another, and wiring a real team together."
read essay →Research with AI #4: The Prompt Is the Work
Five turns before the prompt was right — and the answers ninety minutes couldn't hold.
"Why writing the prompt is itself the methodology in AI-assisted literature review — answers on the reproducibility stack, governance, coverage, and warrant."
read essay →Research with AI #3: Automating My Research (I Wasn't)
What happened when I built my own agentic AI research team
"Everyone's building agentic AI research teams. I tried it too. Turns out I was doing methodology, not automation."
read essay →Research with AI #2: Agents, Honestly
What every AI agent type trades away, and how to choose yours
"AI agents for knowledge workers: every deployment option trades away safety, accessibility, openness, or sovereignty. A framework for choosing honestly."
read essay →Research with AI #1: The Foreclosure Problem
AI makes you faster at finding what you already know to look for. That's the problem.
"The foreclosure problem: how AI literature review tools optimize for speed over discovery, quietly narrowing what you might have considered — and how to build a Claude Code thinking partner that broadens it instead."
read essay →