Kyle Mathews

Big announcement! I'm now a co-founder at @electricsql along with the @thruflo / @balegas and the brilliant team they've built. We're solving local caching of Postgres data. Electric is a VC-backed devtools OSS startup. We're close to our first beta release so check us out!

@Bewinxed haha nice! How often does it find stuff for you?

@kylemathews Boy do I have the library for you github.com/bewinxed/llm-w…

I need some sort of claude code plugin which detects whenever claude writes "pre-existing" and immediately stops everything and sends "THERE ARE NO PRE-EXISTING ISSUES!!!" Opus 4.6 is much lazier than earlier models

@true_droid I think I did actually but it's still happening 💀

@kylemathews Have you tried putting that in your claud.md? I think I'll do that the next time I see him say it.

@threepointone Ugh sorry man.

lemme check twitter before going to bed…

@META_DREAMER @geoffreylitt you can use sqlite — direct support is landing soon github.com/TanStack/db/pu…

@kylemathews @geoffreylitt No PGLite on react native yet github.com/electric-sql/p…

Hot take, grocery shopping makes more sense than travel planning as an AI use case. Frequent and semi-repetitive

@META_DREAMER @geoffreylitt What was wrong with react native?

@kylemathews @geoffreylitt Wow building something almost exactly like this, started with electricSQL + tanstack DB but had to switch to powersync so it could work with react native

@cynthiamcgillis @geoffreylitt It tracks ingredients that we have and don't have and we can mark when we run out

@kylemathews @geoffreylitt How does it know your physical inventory?

@NckLcs @tannerlinsley @trpcio out in 0.0.20

@kylemathews @tannerlinsley @trpcio Mega! Let me know if it would help for me to try a PR release on tRPC before merging!

Generating and shipping skills for your NPM library has never been easier with @ tanstack/intent 🧠 Generate skills - `scaffold` ✅ Validate structure - `validate` 📦 Publish to npm 🔍 Registry auto-discovers! 🔗⬇︎🧵

@NckLcs @tannerlinsley @trpcio got a PR up for this — should be released fairly soon. It turns out scanning for all packages can get tricky haha github.com/TanStack/inten…

@kylemathews @tannerlinsley @trpcio It runs very quickly, but unfortunately also just exits right away whatever I try! Seems to be this issue for us github.com/TanStack/inten…

@kerematam @devagrawal09 @theo oh interesting so you're streaming tokens through a stream and then using the query to figure out where to start streaming? What is a "follow-up" stream?

I am planning to roll out one using Durable Streams + useQuery, where the query returns an offset id, and the follow-up Durable Stream starts immediately after with that offset. It may not count as a full rollout, but I am not planning to use the existing sync engines directly either. I wonder @kylemathews 's input though, would it be just better if i fully commit to tanstack db or electric sql.

I have no idea how people are building reliable AI interfaces these days without a sync engine Please help me understand Or you made the mistake of building your own bespoke sync engine like @theo

@NckLcs @tannerlinsley @trpcio the maintainers SKILL does most of the work!

@tannerlinsley @kylemathews Need to do this for @trpcio but have so many projects on the go already 🥲

@Swizec yessir

@kylemathews like some sort of actor model of computation 🤔

Software 3.0 === multiple stateful entities coordinating in real-time

Check out the new @tan_stack skills registry. It's like `npx skills add` but actually uses a package manager. So your skills actually track your dependencies as they evolve.

@simonw We just shipped this x.com/tan_stack/stat… and for @ElectricSQL x.com/ElectricSQL/st… With them you can now one-shot complex apps which before was heading the agent off from guessing badly at stuff and laboriously feeding in docs

A short note that the predictions that LLMs would favor "boring technology" that's over-represented in the training data don't appear to be playing out as expected with the latest models - once you attach them to a good coding agent harness at least simonwillison.net/2026/Mar/9/not…

@kylemathews the skill: github.com/KyleAMathews/h…

guys stop asking claude for advice and start using @kylemathews dialectic skill for a nuanced deep dive. just had an insightful question on our startup's positioning/initial gtm focus and it was incredibly helpful

@RayFernando1337 @samhogan 1/H sets the ceiling electric-sql.com/blog/2026/02/1…

Velocity with agents is a fallacy right now because orchestration overhead eats the gains. You can ship a lot more if you run agents serially and front-load verification. Building new infra for 1000 concurrent agents assumes the parallelism is worth it. The bottleneck is actually agent reliability and not git.

What if a codebase was actually stored in Postgres and agents directly modified files by reading/writing to the DB? Code velocity has increased 3-5x. This will undoubtedly continue. PR review has already become a bottleneck for high output teams. Codebase checked-out on filesystem seems like a terrible primitive when you have 10-100-1000 agents writing code. Code is now high velocity data and should be modeled at such. Bare minimum, we need write-level atomicity and better coordination across agents, better synchronization primitives for subscribing to codebase state changes and real-time time file-level code lint/fmt/review. The current ~20 year old paradigm of git checkout/branch/push/pr/review/rebase ended Jan 2026. We need an entirely new foundational system for writing code if we’re really going to keep pace with scale laws.