A working notebook · 2026

Experimenting at the intersection of AI, investing, and business / systems intelligence.

Writing about its impact from 20 feet to 20,000 feet. Essays that build what they argue.

RolePrincipal · Queens Court Capital
PriorH.I.G. · Comvest · PGIM
Experiments10+ built, more in flight
§ About this blog

A working notebook for AI, investing, and systems.

Three altitudes, organized by how close to the work you want to be.

Three
altitudes.

§ 01 · The structure

Same shape at every scale.

Each piece of writing lives at one of three altitudes, from specific builds at ground level to structural shifts overhead. Most of the useful work is in the lines between them.

20,000 ftShifts
What changes.

What changes economically and structurally when the cost of producing intelligence collapses, from Brooklyn to Bangalore to Lagos.

Long-view pieces
2,000 ftPatterns
How it works.

How the systems actually work underneath. How cognition maps onto architecture. The structural rhymes between brains, models, and the work people do.

Pattern essays
20 ftBuilds
What to make.

Specific experiments. Real data. Working code. What happens when one person points AI agents at concrete problems and lets them run.

Experiments
§ 02 · How it works

One principal. Many specialists.

One principal directs the work. Stable associates handle writing, analysis, engineering, and audit. A wider specialist bench comes in when needed. What matters most is judgment, agency, and the ability to drive execution.

Separate harness Claude Code Terminal-native execution. Invoked when the work belongs in a shell.
Principal · Orchestration Project Workspace Routes tasks, manages context, and switches the harness to match the work. Sees the full picture and delegates accordingly.
OpenClaw·Hermes·orchestration layer
Separate harness OpenAI Codex IDE-native execution. Invoked when the work belongs in an editor.
AssociateCreativeWriting, design, strategy. Nuanced copy, frontend, visual storytelling.
AssociateAnalysisResearch, fast iteration, data processing. Runs sub-agents and background work.
AssociateEngineeringCode generation, debugging, architecture. Powers 10+ experiments.
AssociateAuditQA, review, verification. Checks that work actually does what it claims.
Model pool · swappable underneath
ClaudeGPTGeminiGrokKimiMiniMax+ local

The firm's infrastructure is OpenClaw and Hermes, open-source orchestration layers. The principal runs the project workspace and switches the harness when the work needs a different lane: Claude Code for terminal-native execution, OpenAI Codex for IDE-native work. Associates handle the standing disciplines; models in the pool are swappable underneath. What remains expensive is judgment, agency, and the ability to drive execution.

§ 03 · Selected experiments

What the system produces.

10+ experiments built from scratch with AI agents across investing, trading, and research infrastructure. Four of them, in brief.

N°·001
Swarm Intelligence

100 AI Analysts Debate Peloton

Structured adversarial simulation. 100 agents with competing incentives. Five rounds of debate. Convergence reveals fragility.

Round 1 — the debate starts dispersed, with only a slight bearish edge.

55% bearish
The Finding

Peloton converged to a strong bearish consensus by round five. The Lululemon control did not converge, which suggests swarms can separate fragile theses from genuinely contested ones.

Gemini ran the agent swarm. Opus synthesized the output. Total cost: a few dollars.

N°·002
Business Intelligence

Chicago Permit Velocity

Permit timing isn't one Chicago market. This tracks where approvals move fastest, where they stall, and which neighborhoods are improving.

Neighborhood intelligence
7.7×
spread between the fastest and slowest permit markets in the same city
Median42d
Ranked75

Fastest → slowest

Citywide median marker at 42 days threshold-adjusted
The Finding

Permit timing is not one citywide market. It is a collection of neighborhood markets with materially different operating conditions. That makes permit velocity useful for underwriting, expansion planning, and local business timing.

Codex wrote the pipeline. Opus audited the methodology. Total software cost: zero.

N°·003
Research Infrastructure

Autonomous Briefing Desk

A recurring research engine for PortCo, AI, and Bitcoin intelligence briefs — source collectors, agent synthesis, signal scoring, and executive-style memo generation.

3
briefing desks
One loop pointed at three domains: collect, filter, synthesize, route.
PortCo Intelligence Brief
OperatorPortCo
AI Daily Brief
Daily
Bitcoin Brief
Market
SourcesSignalsBriefsActions
The Finding

A brief is not just content. It is persistent attention wrapped in a repeatable loop: collect the world, compress the noise, and route what deserves judgment.

The same engine can be pointed at an operating company, the AI ecosystem, or a market thesis.

N°·004
Local Intelligence

Local Market Pricing Intelligence

A competitive intelligence system for fragmented service markets — pricing, review signals, and geographic tradeoffs synthesized into operator-grade local positioning.

Composite local score
0 of 100

Pricing, reviews, and geo stickiness synthesized into one operator-grade local read.

Signal stack
Price positionP78
Review signal84 / 100
Geo stickiness12m
The Finding

Once public pricing is normalized, the market stops looking like isolated websites and starts looking like structure: a floor, a midpoint, a premium tier, and a real spatial competitive set.

Public rates, review context, and basic geo logic become a useful benchmark surprisingly fast.

Other experiments
Consumer Market Simulation · 300 AI personas SPX Options System · Paper-traded learner Market Autopilot · 6 local markets Research Paper Packages · Field dynamics + I-Space
10+ total
§ 04 · Writing

Latest essays and field notes.

View all →