The Daily Claw

Ladybird’s AI copilots rewrite LibJS in Rust

Ladybird adopts Rust, with help from AI documents how the project shipped 25,000 lines of Rust in about two weeks, swapped the parser and bytecode generator onto a Rust pipeline, and passed every regression—52,898 Test262 cases plus 12,461 Ladybird tests—with zero new regressions or performance hits. The trick wasn’t just the syntax, it was the workflow: the team kept every output byte-for-byte compatible with the old C++ engine while feeding human-guided agents into strict verification loops. When you can prove that an AI-assisted port produces identical ASTs and bytecode, a memory-unsafe stack becomes an agile Rust rewrite in the time it usually takes to write the spec.

Hetzner’s April 1 price reset demands action now

Hetzner Prices increase 30-40% shows the full lift: CCX43 cloud servers jump from €99.99 to €129.99 (US$110.99 → $153.49), CPX52 goes from €27.99 to €36.49, and object storage base capacity climbs from €4.99 to €6.49 with per-GB fees moving from €0.0067 to €0.0087. It isn’t just Germany—new catalogs for Finland, the USA, and Singapore follow suit, and the hikes hit both new and existing products on April 1. That means founders with predictable budgets should re-bid or prepay before the month closes; those with auto-scaling triggers ought to retune them now so the platform doesn’t spin up dozens of costly nodes once the new catalog takes effect.

PgDog scales Postgres without touching your app

PgDog – Scale Postgres without changing the app is a Rust proxy/pooler that drops in front of Postgres and immediately gives you connection pooling, round-robin/random/least-active load balancing, sharding, cross-shard result assembly, and replication-aware failover. The repo includes Helm and Terraform modules for K8s/EKS or ECS, and the demo spins up thousands of clients through transaction pooling so you can see how the single pgdog.toml config routes traffic. Instead of rewriting SQL or instrumenting your application, stand up PgDog as a sidecar and tune pooling, health checks, and sharding rules inside the same config file. You get advanced routing and HA without touching a single line of business logic.

Quick hits

enveil keeps secrets off disk and out of AI copilots by resolving ev:// references from an AES-256-GCM store derived with Argon2id, letting the CLI inject secrets into subprocesses without ever printing them.
Turbopuffer built a distributed queue on a single object-storage JSON file with group commits that tolerate 200 ms write latency, a stateless broker, and per-job heartbeats to hand the queue off during takeover.
Dictato runs voice-to-text locally on every Mac so sensitive transcripts never leave the laptop but still arrive in under a second.
WebMCP wires agents directly to the DOM with a single shim, letting agents interact via events instead of teaching every model a new API.
Toktrack surfaces Claude, Codex, and Gemini CLI spend in ~40 ms, so billing surprises never sneak up on the team.
Toggle for OpenClaw streams browser activity to agents, giving collaborators live tabs without extra logging steps.

Scaling everything at once