UNIT 01
Signal Forge
AI-powered trading rule evolution. Karpathy's autoresearch pattern applied to options signals — the loop optimizes itself overnight.
v6.9 LIVEIssue 047 Featured Filed 2026.04.25
From compute explosions to org hierarchy collapse — what my research vault is telling me about the next 18 months. Plus one non-obvious conclusion about enterprise software.
I spawned 13 parallel Claude Code agents on a Saturday morning. Each shipped one feature for my trading system. Agent 12 overturned the hypothesis from Agent 1.
88 concept articles, 301 sources, 13 agents. Most "second brain" stacks fail because they confuse output with compounding. Three Laws decide which yours is.
Animal names are debt. Functional names are leverage. The rename took 30 minutes and touched 26 files. Layer 1 vs Layer 2, and why naming is plumbing.
A 90-minute lint pass: 19 missing banners, 132 unindexed notes, 24 confidence gaps, 6 open concept gaps, plus a contradiction scan. Whole pass, full receipts.
A 90% winner with a −1.04% expectancy. The Trojan Horse hiding in every sales pipeline — and the five-minute math that surfaces it.
A 3-weekend prospecting stack that replaced three subscriptions.
Compiling a personal wiki so Claude answers from my actual notes — not training data. Karpathy's autoresearch loop, applied.
Four tools. Under $200/mo. Replaces what most AEs pay $2,000/mo for. The math, the workflow, the gotchas.
ServiceNow's token tax. Sierra's per-resolution model. The pricing posture every AI vendor is converging on — and the one that breaks the seat license.
AI-powered trading rule evolution. Karpathy's autoresearch pattern applied to options signals — the loop optimizes itself overnight.
v6.9 LIVEWeekly 4-agent pipeline on top accounts. Intel → triggers → angle → message.
ACTIVEAlways-on personal AI agent. Slack interface. Memory. Cron. The OS layer for everything else.
50% BUILTVerification layer for AI agents. Evals as the new moat — the skill nobody teaches.
SPRINTI write about the infrastructure work between a chip and a quota — and I ship the AI tools enterprise GTM teams actually need.
Twenty years in enterprise software sales — Oracle and Salesforce in earlier chapters, most recent stint independent. Now building AI tooling for sellers, evaluators, and operators. Based in San Francisco.
This blog is a build log. Every post is a problem I hit, a tool I shipped, or a thesis I'm pressure-testing in the wild. No daily digest. No reposts.
One issue per week. Long-form essays on AI infrastructure, evals, enterprise pricing, and the reliability work between a chip and a quota. No daily digest. No reposts.