Hey there!
Summary of this email:
- New edition of the AI Workshop for devs: November 26, only 20 seats
- Why not use Jev to search your database?
- Café con Codely recap (watch it here)
- The joke of the week (database themed!!!)
🤖 New edition of the AI Workshop for devs
We have opened a new edition of the AI Workshop for devs. We added the most relevant concepts for building products today.
It is on Thursday, November 26, in Spanish, and there are only 20 seats.
🔍 Why not use Jev to search your database?
A few weeks ago pg-jev went viral. It is a Postgres extension that adds a jev() function to search your database in natural language, without indexes and without embeddings.
The demo was very tempting. They had a table with 129 rows and, inside the WHERE, they asked Jev to keep only the ones that match a condition written in natural language. Jev checked the 129 rows in about one second for $0.0009, and the second time it took 6 milliseconds thanks to the cache.
But there is a catch. To filter, pg-jev sends all the rows to Jev in groups of 20 and asks which ones match the condition.
The problem is that this does not scale. If you want to filter a table of one million rows, each search takes almost half an hour and costs $6.20.
The alternative that does scale is to use an embedding search. With embeddings, you pay the cost only once, when you insert the row, and the search is a math operation inside the database.
With one million rows of about 50 tokens each, indexing everything costs about 60 cents and each search is practically free and responds in milliseconds.
That said, pg-jev wins in one thing: it understands conditions. If you search for "running shoes that are not Nike", embeddings return exactly the Nike ones, because the word Nike is in the query. Jev understands the "not".
Our take is simple: to search, use embeddings. If you need to rerank the results, that is where you can use Jev.
We explain all of this step by step in the Jev and System One models course, which is now 100% published. 😊
☕ Café con Codely recap
This is the top 3 of headlines we discussed in the latest Café:
- Rex, the Superlogical terminal: persistent sessions, native Tailscale and tabs on the left. The desktop app is only a client of a server, so you close it with Cmd+Q and your sessions and SSH connections stay alive.
- Spotify: AI does not lower code quality, but the processes must adapt. They went from about 8,100 PRs in August 2025 to 17,000 changes in August 2026, with more incidents that do not come from worse code.
- The bomb on the bots: what goes down is Yandex, not the social media bots. In the Cloudflare Radar charts, "bot" means automated HTTP requests, like the ones from the Yandex crawler.
You have the full Café recap with links to each story at the moment we discussed it.
Tomorrow we will be live at 9 CEST discussing this week's news on Café con Codely. On our YouTube, Twitch and X. 🙌
And since you've made it to this part of the newsletter, here's the joke of the week, which I know you were waiting for:
Why did the database admin break up with their partner? Because the relationship was one-to-many! 😂 😂 😂
Cheers!