01
Teardowns
Honest dissections of our own tools. What it solved, where it lied, what had to go. We started with the Stock Analyzer.
Finance runs six months behind code — and it's taking the same road. The work is moving off spreadsheets and onto agents: skills, connectors, subagents. The edge isn't speed; it's architecture, data integrity, and knowing exactly where the tool breaks.
This section is a finance engineer's notebook. We build real tools, then leave them right where we tore them open. No marketing — post-mortems.
Honest dissections of our own tools. What it solved, where it lied, what had to go. We started with the Stock Analyzer.
The architecture of a financial workflow: skill / connector / subagent, orchestration, data integrity, and why human sign-off isn't negotiable.
A vendor-neutral map. What Claude, Gemini, OpenAI and specialist agents actually do in finance — benchmarks and governance included.
End-to-end, reproducible build logs. The newsletter pipeline, the research desk, the job-hunter, FP&A models.
The first field note that names exactly where the tool broke while building a stock-analysis pipeline.
● LiveHow we caught "made-up" ETF-flow, funding-rate and open-interest numbers, and rebuilt data integrity.
● LiveVals AI and Forrester benchmarks, MNPI and governance notes — an architect's read, not a vendor pitch.
● LiveThe reference architecture is three parts. We open each one with examples from our own pipeline.
Building the subscriber system from scratch: architecture, the double opt-in flow, and shipping it live.
Wiring multi-source ATS polling to a cron: design decisions and maintenance cost.
Finance is the next stage for generative AI. Below is a vendor-neutral frame: where each player sits in finance.
| Player | What the finance side ships | Where it's strong |
|---|---|---|
| Claude | Claude for Financial Services: 10 ready finance-agent templates (pitchbook, KYC, month-end close), MS365 add-ins (Excel/PPT/Word), a Moody's native MCP app. | Agent architecture (skill/connector/subagent), long-document and model work, auditability. Leads the Vals AI Finance Agent benchmark. |
| Gemini Enterprise | Financial-analysis templates in Agent Garden, Gemini in Sheets, partner agents (Kensho/S&P, AutoCIO, Obin, Genpact). | Low-friction inside Workspace/Sheets, large-document summarization, governance and ecosystem integration. |
| OpenAI | Token-based workspace agents, fast prototyping, a clean API. | Customer-facing flows, quick pilots, strong on complex financial modeling. |
| Specialist agents | Kensho/S&P data retrieval, AutoCIO portfolios, Obin private markets, Moody's credit/risk. | Data-adjacent, going deep on a single vertical; the layer that feeds the general model through a connector. |