Home sections, draft v1
What I'm building now
- An always-on AI agent on a Mac mini: Claude Code on a schedule, my phone as the control channel, and a separate critic that checks its work.
- This site: The Denominator, built on EmDash, with notes on running AI agents in production for data teams.
- A Healthcare / Life Sciences dataset for the Snowflake Marketplace. [BEN: name the pick once decided; Ray's current pick, awaiting your read, is a Medicare coverage-policy knowledge extension for Cortex.]
- Certifications as proof of homework: Claude Certified Architect – Foundations (August 2026) and SnowPro Specialty: Gen AI (September 2026). [BEN: show scores 850 / 854, or not?]
- [BEN: optional fifth line, e.g. a small semantic-view demo on synthetic patient data for post #1's build slot, only if you plan to build it.]
Reading list
Five things I'd hand a data leader who is about to put agents in production.
- Building effective agents, Anthropic. Separates fixed workflows from agents that direct their own steps, and argues for the simplest option that works.
- Effective context engineering for AI agents, Anthropic. Treats context as a limited budget: find "the smallest possible set of high-signal tokens that maximize the likelihood of some desired outcome."
- Context Rot, Chroma. Tested 18 models and found performance drops as input grows, often unevenly. A good reason to stop pasting everything into the prompt.
- Semantic Layer vs. Text-to-SQL: 2026 Benchmark Update, dbt Labs. A vendor benchmark, so read it with that in mind; its useful finding is that a semantic layer fails with an error instead of a plausible wrong number.
- Apache Ossie (Incubating): The New Name for Open Semantic Interchange, Snowflake. The Open Semantic Interchange spec, now Apache Ossie: one vendor-neutral format for metrics and dimensions across tools.
Ben to decide
- Mac mini detail on a public home page: keep, or say "a small always-on machine at home".
- Dataset line: hold it until the pick is made, or keep it vague ("a Healthcare / Life Sciences dataset").
- Cert scores on the home page.
- Reading list item 4 is dbt Labs' own benchmark of its own product. Keep with the "vendor benchmark" flag, or swap for Anthropic's multi-agent research system post (also verified working).
- Reading list item 5 is a Snowflake blog post. Alternative for a practitioner voice: Yannick Misteli, "The semantic layer is not a layer" (verified working).
URL verification (2026-10-02)
| Item | Check | Result |
|---|---|---|
| Building effective agents | curl + WebFetch | 200 (the vault's /research/ URL redirects here) |
| Effective context engineering | curl + WebFetch | 200; quote confirmed |
| Context Rot | curl + WebFetch | 200; 18 models confirmed |
| dbt benchmark | curl + WebFetch | 200; published Apr 7, 2026; finding confirmed |
| Snowflake Apache Ossie post | WebFetch OK (Jul 8, 2026); curl returns 403 (bot block) | Treat as working; click once in a browser before publishing |
| Multi-agent research system (alt) | curl | 200 |
| Misteli (alt) | curl | 200 |