Thinking in Loops: Finding the Frontier Without a Frontier Budget
Three days into scoring a legal benchmark, my tokens ran out. The fix wasn't a bigger budget or a better model — it was a different shape. On orchestration, loops, and reaching the frontier of what AI can do without a frontier budget.
My Agent Did the Legal Work. The Benchmark Gave It Zero.
Somewhere in my benchmark results sits a complete legal memorandum. Fifty-eight kilobytes of finished work: issues identified, analysis done,
Commentary: The Pitfalls of Seeking AI Advice From Lawyers
A sharp CNA commentary argues AI can't replace lawyers. It's aiming a very good argument at old technology — and missing the real threat.
The Judgment, Not the Summary: How Zeeker MCP Can Change the Way you do Legal Research
Singapore's Employment Claims Tribunal was built to be simple enough that lawyers weren't needed. The first published ECT judgment runs to 414 paragraphs of dense legal reasoning. Here's how to query it in plain language.
Lawyers Are Building. Just Not On Each Other's Code.
Mike got 2,657 stars in eleven days and six merged pull requests. The LegalQuants directory has 112 lawyer-coders and one repo that cleared the bar — mine. Legal open source isn't a community yet. It's a federation of solo-author archipelagos.