Notes
- NOTE Typically in dbt, when we want to deprecate a model, should we remove dbt .sql file and the underlying table entirely or just dbt model, keep table?
- Asked in Ampcode:
- Phase 1: deprecate in code.
- Phase 2: remove the
.sqlfile. - Phase 3: drop the table.
- Why not skip straight to dropping? In our setup, AML models reference BQ tables by physical name. If you remove
.sqlfile, dbt stops refreshing the table but it still exists with stale data, which is actually worse than dropping it, because consumers silently get stale results. - So either keep the wrapper alive or drop the table entirely. Don’t leave in the middle.
- Asked in Ampcode:
- NOTE Prioritization: enterprise customers are considered higher, need to treat them carefully.
- Seems I don’t use the Gemini chat bot anymore. Plan to downgrade the Google One?
- My AI Adoption Journey
- “I quickly discovered for myself from first principles what others were already saying, but discovering it myself resulted in a stronger fundamental understanding.”
- That’s correct. And applies for everything. Learn it yourself result in your own understanding.
- That relates to many people find more interesting reading books instead of watching transfered movies.
- “Very important at this stage: TURN OFF DESKTOP NOTIFICATION. Context switching is expensive”
- That’s correct. From my experience working with many projects at the same time, context switching is the most time wasting and could lead to my headache more frequently.
- New term: Harness Engineering
- https://media.secondbrain.lelouvincx.com/2026/04/e654a15e5b7153246b3280ec13b58ed1.png
- AI is like a horse. It’s strong. Has power to transport things. But without harness, it does not achieve the best effectiveness.
- Harness Engineering in AI means design the better feedback loop so that the AI knows it’s wrong.
- The idea is that anytime you find an agent makes a mistake, you take the time to engineer the solution the the agent never makes that mistake again.
- The simplest form of this is adding rules into
AGENTS.md, that said, I’m already know about this fact, and actively improving this (logseq automation).
- The author: Mitchell Hashimoto
- He is co-founder of HashiCorp, one of the known Infrastructure as Code - Terraform.
- Pretty well-known.
- He created Ghostty as a side project. Pretty well-known terminal emulator (similar to Alacritty which I use).
- “I quickly discovered for myself from first principles what others were already saying, but discovering it myself resulted in a stronger fundamental understanding.”
Tasks
- ((69de5ae8-505b-47ce-a301-f93766780b85))
Partly today
Done
