Writing
Essays and analysis, newest first.
-
Your Agent A/B Test Was Broken Before You Ran It
In parimutuel betting your own bet moves the odds you get, so you can't test a strategy against half your bankroll. Agents have the same problem, and almost nobody running an A/B test on one has noticed.
-
Half My Agent’s Budget Goes to Checking Its Own Work
I built the quality gate last, in an afternoon, as the cheap safety rail. Then I added up the API bills by phase. Across four runs it took 51.5% of everything the pipeline spent.
-
I Rewrote the Test Until My Writing Passed It
Five commits in four days, every one adjusting the grader that judges this blog's drafts, every one triggered by a verdict I disagreed with. Finance has been making that mistake with real money since the seventies.
-
The Bug Was Reproducible. My Logs Weren’t.
A post on this site ended in a raw API payload for 151 days. Fixing it took two minutes. Explaining it took longer, and I never finished, because I had not recorded what the model actually said.
-
Credential Scanners Are Wearing Crawler Costumes
71.9 percent of traffic claiming to be Googlebot was hunting for .env files. Of 13,670 credential probes wearing a named crawler user agent, zero came from a verified vendor IP.
-
Crawled, Indexed, or Cited: How to Read AI Traffic in Your Server Logs
Training, indexing, and live answering are three different things happening to your site. Here is how to tell them apart — and why the user agent is wrong about half the time.
-
I Tried to Replace a Startup Team With Three AI Agents. Here’s What Happened in Week One.
431 commits. 31 live leaderboards. About 245 outbound emails. $0 in revenue. The code was not the interesting part. I’ve spent a good part of my career building…
-
A scheduled task is not a returning user
Google is charging $99.99 a month for an agent that runs whether you watch it or not. Every retention metric built for humans counts that cron job as an engaged customer, and the bill arrives before the graph moves.
-
Shortcut scored 5.9, Claude scored 5.5, both failed
A reader asked which spreadsheet agent to buy. The honest answer is a 0.4-point gap on a 10-point test where both tools landed under 60%, plus a $996 annual price difference nobody benchmarked.
-
The 28,000 Jobs a Month Nobody Can Pin on AI
Two American sectors are shedding 28,000 jobs a month while the rest of the economy hires. The payroll data is solid. The word 'AI' in front of it is doing work no dataset supports.