
My overriding thought going into 2026: Problems you thought were out of reach no longer are. Pick a problem 100x harder than you thought possible to tackle alone, and run with it. Start now. AI will catch up and make it possible.

Building Realtime @neondatabase & @databricks. Prev co-founder @ElectricSQL (acquired by Databricks). Started PGlite.dev, co-creator of @tan_stack DB.

My overriding thought going into 2026: Problems you thought were out of reach no longer are. Pick a problem 100x harder than you thought possible to tackle alone, and run with it. Start now. AI will catch up and make it possible.

Introducing Jerry! - Your personal desktop explorer and friend. - Give him missions on your computer and watch him tackle them. - He keeps a fun journal from his missions. - JOL (Jerry On Line) gives you remote access with live updates via @neondatabase realtime.

What ridiculous, completely unproductive, but FUN thing are you building this weekend? I've given Jerry a torch for dark mode 🔦 (p.s. you can download a ready to run build of him on GitHub!)
This weekend, if you are vibe coding, don't build something productive, don't built a tool for yourself to use, don't build something because you need it. Build something RIDICULES because you CAN! Enjoin the unlimited & unbounded creativity we have been gifted!

Super fun.
Introducing Jerry! - Your personal desktop explorer and friend. - Give him missions on your computer and watch him tackle them. - He keeps a fun journal from his missions. - JOL (Jerry On Line) gives you remote access with live updates via @neondatabase realtime.

Fantastic example of realtime power and personal fun software
Introducing Jerry! - Your personal desktop explorer and friend. - Give him missions on your computer and watch him tackle them. - He keeps a fun journal from his missions. - JOL (Jerry On Line) gives you remote access with live updates via @neondatabase realtime.

The @MubitLabs team built an agent for cutting frontier-model token spend They tried it on an example template, and it cut frontier-model tokens by 73% and ran 40% faster, without losing accuracy Running on Lakebase Postgres, Neon Functions, and AI Gateway

This weekend, if you are vibe coding, don't build something productive, don't built a tool for yourself to use, don't build something because you need it. Build something RIDICULES because you CAN! Enjoin the unlimited & unbounded creativity we have been gifted!

This was my ridiculous weekend project!
Introducing Jerry! - Your personal desktop explorer and friend. - Give him missions on your computer and watch him tackle them. - He keeps a fun journal from his missions. - JOL (Jerry On Line) gives you remote access with live updates via @neondatabase realtime.

You can get Jerry from my GitHub github.com/samwillis/jerry I'd love to see the missions you give him!

When he's not got anything to do he explores your desktop looking for your missing socks.

You can also control him just like a platform game! Walk, run, climb, swing, and even parachute across your desktop.

If coding agents are really as good as everyone says, it makes sense to try again at the intractable problems in databases.
There are two unclimbable mountains in database world: 1 Combined transactional/analytical processing 2 true multi-master scale out We are making significant progress on both in Lakebase

Transactional/analytical is being solved with LTAP and we will swing at multi-master super hard. I'm excited to be able to solve such long standing database problems.

There are two unclimbable mountains in database world: 1 Combined transactional/analytical processing 2 true multi-master scale out We are making significant progress on both in Lakebase

Neon is doing “git for infrastructure”. One branch containing your functions, buckets and database is the way to go.
We ship updates across the Neon backend every week. Recently landed: [Object Storage] If you roll back a branch, your files roll back too [Neon Functions] 24 ready-to-use templates [Lakebase Search] Full-text search in Chinese, Japanese, Korean [AI Gateway] Free plan users can buy credits

If used to be "the best tool for the job is the one you know", that's completely changed, and with it the way we think about a lot of things. So many abstractions were about providing a layer for humans so that their knowlage transferred. Again not really needed any more. I'm increasingly convinced that the right tool for the job is the one with the fewest abstractions over the platform. Building a mobile/desktop app, do it in the OS native toolkits. Building an api server, go for a low level language that scales. The one I'm still trying to work out is websites/apps, what we should be doing there. The frameworks fulfill a vital role on the client, but I'm much less convinced on the server. I probably want a lower level, high performance language on the server. I think we're going to see a new generation of full stack frameworks that use JS on the client, but Rust or Go on the server.