alias

May the bridges we burn light the way.

Sure, your competitors all suck. No doubt about it. You manage the open source databases better than they do. However, PlanetScale hasn’t built a single storage engine. MySQL and Postgres handle anything heavy over there. Vitess was built at YouTube by Sugu and Mike S. PlanetScale contributes to it, sure, but it’s a clustering system built around MySQL. Neki is your one novel piece of work, a distributed query/sharding layer for Postgres, but still built on top of an existing database engine, and it’s closed source. You stand on the backs of OSS giants and then tweet shit like this. Fine. Your business model, whatever… but you’re a managed infra company, not a database company.

If I give my time to contribute to an OSS repo you maintain and you're rude to me; if you make me feel like I need to beg or 'pay for support' for my code to be merged... you're in the danger zone. I am not going to go for it and I'm going to apply an undue amount of effort to make you feel that.

GitHub is on its last legs in more ways than AI. They’ve lost the plot in their core business model.

I have a feeling this is going to be increasingly common.

I remember everyone saying that Elon couldn’t just rewrite it as if it was somehow this impossible task. Haha.

You guys gotta give us a Rust version.

We’re going to get a shitload of formal verification startups and crappy dst engines.

How are you using it? Did you just replace Codex or Claude or whatever else you’re using with their harness and use frontier models or whatever else?

I think Apple did this, didn’t they? Wasn’t this an update the pushed that rolled them back? Foolishly.

This is the first thing I’ve seen in like two years that gets me actually excited.

Alright, this is cool. Is there any source we can review? Is this C only?

This doesn’t make sense to me. Lean4 is great for one thing - formally verifying math. Outside of that, it’s not super useful as a programming language, IMO. TLA/TLC/Apalache are obviously not going to get it done. Verus is built around Rust, and isn’t going to cut it. Maybe Bend2 can work, but it’s the first swing at the problem. Granted, it’s incredible, but I don’t think we’re going to see all of us just stop writing code all together. Rust is the only thing that matters anymore, IMO; at least outside of Bend2 or languages we don’t know exist yet.

Excellent use of free will. I do wish you’d chosen a different name, though. You’re certainly aware of Zed, the IDE. I feel like they’ve earned the right to a monopoly on the name. I assume you’ll open source this… which makes it equally frustrating for both your work and theirs to get proper attention. I don’t know. Either way, this is exciting.

Wire protocols aren’t some hyper optimized choice to improve the UX. They’re job security and ecosystem lock-in for the wire protocol. I fucking hate them - all of them.

Dude this is gold. You should push.

You have great taste in like quick and dirty tooling builds… but I have too many questions to use this one.

I’ve spent two years iterating on the product that solves this problem. At least I know who to send the first demo to.

SQLite kind of cheats to get here relative to consistency, but yeah… it’s fucking incredible.

I’ve read it. I’ve actually written something kind of similar a while back. If you’re interested, it’s here: loadingalias.dev/posts/git-w… I feel like you and I should talk more. Haha. We keep meeting up in random Twitter replies. I’m not going to write a book, but I definitely have opinions. I think the simplest way to state them is that any system whose foundation was built non-modularly post-agentic development isn’t likely to cut it. I have followed ERSC, what little is publicly available anyway, for a bit. Maybe it’s better to say I was aware something was going on there. I know the brain power there is serious. I just can’t help but feel like it’s missing the bigger picture. A new storage engine is an awesome start - a critical one, even - but JJ doesn’t really solve anything outside of a slightly better DX as of now. I hope that changes, but we’re still not talking about doing less work, about breaking down some of the fragmentation that’s calcified across VCS in the last decade. We’re still not talking about agentic history, commutative edits, or chunking on the host (any new VCS needs to move minimal data across the wire via some kind of CDC + QUIC/IBLT handshake or something like that - idk). What do we do about the everyday engineer who is using worktrees to determine what needs to be implemented, like a real-time A/B, where history from the agents in all worktrees is relevant to future work? What about his teammates? What sort of storage engine are we talking about? I assume you’re a Rust shop - only because I know your work there - what sort of SMR scheme is allowing memory reclamation at today’s scale? How do we shrink the command surface for humans and reduce the resource consumption for agents who don’t need human level commands? I guess I feel like it’s along road. I’m really rooting for ERSC - as of now… it’s the only game any of us take seriously in this respect.