Weekly notes: Getting back into reading Seth Godin’s blog, llm context vs prompting and a few other things to round out the week :)

A Deep Dive into Solid Queue for Ruby on Rails: Interesting architectural overview of of a job queuing system.
Scaling our Observability platform beyond 100 Petabytes by embracing wide events and replacing OTel: Internal ClickHouse central log system and strategy. CH looks like a solid choice for an observability database.
Mud, water and wood: The system that kept a 1604-year-old city afloat: Hundreds of thousands of wood spokes are pounded into the earth underneath Venice. It has stood for 1600 years and will keep going. Buildings today are built to last 50 years or so.
The Balatro Timeline: Developer diary for one of last year’s most popular indie games.
Constraints naming conventions (Database): How gitlab thinks about naming tables, keys, and indexes.
- pk_table_name
- fk_table_name_column_name
- index_table_name_on_column_name
The New Skill in AI is Not Prompting, It’s Context Engineering: Different framing for how we get llms to string together more relevant sequences of word tokens. It’s less about the immediate prompt and more about all the information you feed into the llm before you ask that supports your question.
Three choices: Seth 3 of the biggest choices you make when you’re getting started on a project. Landlord, investors, customers. Kinda nice framing.
Serving 200 million requests per day with a cgi-bin: Interesting thought exercise here about how much performance you can get out of modern hardware using the old “heavy” process-per-request model of handling transactions in a service. He’s using golang which helps. (Static linkage, faster startup times, and efficiency ftw)