I’m starting a new company with few good people and an exciting idea. I appreciate all the support. More updates, summaries, feels, etc will follow.

Authors sometimes talk about how certain characters insist on doing certain things and they have to go along. I never really understood it. But now I have a presentation outline. And I'm filling in the details. And the story in my head refuses to match the outline.

America needs you! Join U.S. Immigration and Customs Enforcement today.

My favorite type of paper isn't "Here's a brilliant algorithm for solving a very hard problem". But rather "Here are 10 techniques we used to do something really challenging in production". The former is intellectually satisfying. The latter is more likely to be useful.

Importantly, this method is more robust than mere replication. The fact that these salt-seperated duplicates are indeed duplicates isn't known to any part of the system, and so there's no way their durability or availability is correlated.

The centaur / reverse centaur thing feels particularly useful to me Centaur = a human that gets to control AI to enhance their skills Reverse centaur = an AI system that gets to direct and control a human

Market turbulence calls for balance. Gold could help your portfolio hold steady.

This snippet from “on-demand container loading in AWS Lambda” paper by @MarcJBrooker et al, reminded me of a certain storage vendor who tried to sell us on their block-level dedup technology. By highlighting how much money we will save if we dedup our production database backups.

I managed to trigger some Gemini bug…

"PostgreSQL as a JSON database: Advanced patterns and best practices". https://t.co/2NEkfwlTWq

For a new project I'm working on, I wanted to do a comparison between the price/performance of Digital Ocean, #Vultr, and #Hetzner. Unfortunately, DO decided to block me for unknown reasons and their support is non responsive, so I ended up comparing only Vultr and Hetzner:

The event of the year :)

@gwenshap Another alternative is to store the ordinal value while the column has a FK to a description table. This design allows the Enum column to be very compact without sacrificing descriptiveness (we can join the description table) or data integrity (FK constraint allows only valid

Our average cost per month for CrowdHealth hit its lowest EVER in 2025!!! $140.42/month for singles <55 (down 13% vs '24) $225.83/month for singles 55+ (down 17% vs '24) $476.25/month for families of 4 (down 11% vs '24) If you are looking for an affordable alternative to health

https://t.co/YWMQptlAqx

We chose PostgreSQL over MongoDB for our analytics platform. The context: - 50GB of time-series data daily - Complex queries with joins across multiple dimensions - Team had more SQL experience than NoSQL MongoDB seemed obvious for scale, but: - Query complexity made

Do people still do multi-shot prompts? Or did better LLMs made this practice obsolete? It used to be so common. I haven't seen it mentioned in ages but maybe its still quietly useful?

So cool. Like Postgrest, but in Go rather than Haskell. And with pluggable custom routes!

Santa’s gone digital. Get your personalized Cameo from Santa, and give the special ones in your life a surprise they’ll never forget.

I just brutally cut another 3rd-party dependency. Our Vector Database: Pinecone. Pinecone used to be an essential building block in Aidbase's RAG system. It would store tens of millions of individual knowledge points from the custom models our users are training on their

Rule of thumb for any established business: they know what converts. And who they want to reach. If you don’t like their approach to pricing, you’re actively being repelled. They don’t want you to use their service. It’s intentional. You’re their Non-Ideal Customer Profile.

Overheard: "The thing about congestion control algorithms is that most of them don't work"

Just read Anthropic's case study on building long-running agents. Two key takeaways: - AI engineering is still fundamentally prompt engineering. Even for agents. Getting prompts right for each step matters more than complex workflow architecture. - They instruct the LLM to call

.@KimStrassel: “I shiver to think about how close we came to full-on censorship under this last administration.”

If you are interested in Postgres internals, C and Posix, this is an amazing talk. Really highlights key details and complexity around threading and process models in C (in its various iterations).

NGL. Seeing @confluentinc logo on Hadjar’s wheel feels me with joy every time.