Patent systems are first-to-file. It's produced a quiet two-tier system: well-funded companies file in 48 hours. Everyone else waits weeks. Today we're announcing @fearn_ai's $5.5M seed to end that gap. 🧵

If you've been wondering what I've been up to, the team and I have been cooking up something new. A new agent-native development environment deeply integrated with the GitHub graph. Not just for writing code, but all of the meta-work as well.

If you have a project that needs fast, reliable, git based code storage that is built with modern, agentic workflows in mind then we should have a chat about code[dot]storage DM here or email kris[at]pierre[dot]co

I thought I provide a counterpoint of what Tauri provides, as we migrated our Electron app to Tauri. As with any technology, everything is just full of tradeoffs. Since Tauri has a rust core, locks, threads, shared runtime memory, and os calls are much more performant than javascript/typescript. We heavily leverage concurreny, as we read and write to files on disk. Using a system language is going to be far more performant when manipliating a filesystem. Tauri is also inherently secure out of the box. Tauri explicitly requires you to grant permissions for the filesystem, network, etc., and lets you scope resources. With Electron, you do have to open a port to communicate with the Node.js runtime, which might be a feature for some people. Tauri uses native bridges and function closures between the webview and Rust core, which is inherently safer than local-network IPC (you can encrypt data between the network calls, but you pay for that in performance) as the data never leaves the runtime. Also, using the OS webview, your app benefits when users update their OS, and the binary is smaller because you do not have to package Chromium separately. But the real reason we migrated: native libraries. Using libgit2 via FFI and jj-lib natively is worth the tradeoff of WebKit bugs. We pass an OAuth2 token at runtime and run git push / git clone under the user's own credentials. Yes, we do face painful WebKit bugs, especially around rendering the text cursor. But we thought that the cross-platform bugs were worth the trade-off of the security benefits. I do believe that you can make an Electron app just as secure by using service workers and iframes. It is more work, and you do give up performance for it. But you can do it. If you want to go deeper, I gave a talk at Git Merge: youtube.com/watch?v=4UVzX3UF… that covers our migration in more detail.

I am trying to build the fastest way to query a CSV with SQL. I have it down to 2 clicks.

I've turned this into a blog post with an interactive visualization: "TurboQuant, or: why are random compression schemes often nearly as good as carefully-designed ones?". Link in reply.

Alex is bringing markdown to life and it can't happen soon enough.

we live in a wild time. I was able to write a grpc tool, with reflection, in an afternoon. this would normally have taking me at least a week with grpc finiky bs. this also supports tls.

I visited the Pierre office and can attest that the vibes are spot on and they are working on some hard shit. Their plants and speaker game is very strong.

"Run your entire business out of a markdown file", aws COGS edition. Look. When I search resource list for `gitea`, the total/forecasted costs should update too, so that I just see the costs for `gitea` resources. This is my #1 request for @JoinVantage. I got tired of waiting so I built a template on top of the API (see next tweet).

Ok. For the last, like, 15 years, I've constantly been told that CRDTs (and Yjs in particular) solve collaborative text editing. Both the offline and live-collab cases. After an excruciatingly painful evaluation, I argue not only is this completely false, they're generally not appropriate for production-grade editors... at all. Yeah, I know: everyone uses Yjs, so the problem must be us. Right? That's what I thought too. Well, did you know that on every single collaborative keystroke, Yjs will REPLACE YOUR ENTIRE DOCUMENT? (See this GitHub issue: github.com/yjs/y-prosemirror…) Did you know that this is BY DESIGN? (See this discussion on the ProseMirror boards: discuss.prosemirror.net/t/of…) Did you know that this breaks, like, every editor plugin? (See the ProseMirror author's commentary here: discuss.prosemirror.net/t/of…) I am not trying to bag on anyone, but to me this kind of mistake belies a fundamental misunderstanding of what text editors need to perform well, in any situation, let alone a collaborative one. Anyway, learn from my pain. I wrote a somewhat long-form article about all the other challenges we ran into with CRDTs on our blog: moment.dev/blog/lies-i-was-t…

We wrote about our live collaboration solution and why a centralized server approach worked for us vs. CRDTs. If you are interested in collaborative editing, this is a must read!

We have been working hard to launch our take on collaborative editing, as it looks like it is in the new agent world for developers. We believe it looks multiplayer, with seamless integration with git (secretly it’s a mix of jj and git!) and live collaborative editing. We would love for you to try it out and hear your thoughts!

github, this is bullshit

Working with agents reminds me of working with 3D renders. You set a bunch of parameters and input data (model files, light positions and settings, shader files/code, point files for environments, etc., etc.) and send it off (in my line of work, it was sending it off to a render farm) and come back the next day and see what happened.