I don't have time to build these things, will you?

TL;DR

• 1

  Rebuilding NPM is suddenly plausible: Theo argues package publishing, revocation, auditing, metadata, name squatting, and NPX safety are all broken today, and points to Socket catching exploits before NPM itself as proof there is room for a better registry.

• 2

  Git's repo-level permissions are a design failure: he wants source control that can hide .env files, private branches, in-flight PRs, and security fixes, because the current all-or-nothing model makes open source, secret management, and coordinated vulnerability disclosure much worse.

• 3

  File systems are becoming the bottleneck for agentic dev workflows: Theo cites a benchmark where a cached install took 6.8 seconds on Ubuntu versus 31 seconds on an M4 Mac, with some Apple setups reaching 140 seconds, and uses that to argue for in-memory, file-system-light tooling.

• 4

  A 'Dropbox for devs' could solve cross-machine chaos: his vision is one synced code directory across Macs, Linux boxes, NAS, and cloud agents, with lazy-loaded files, shared structure, and environment state that follows you without Git submodule hell.

• 5

  Mobile lost great builders because app distribution is miserable: Theo ties CyanogenMod, Paranoid Android, BlackBerry 10, and Paul Henschel's path from ROM hacking to Zustand and React Three Fiber into one claim, that building or modifying the platform used to feel easier than shipping apps on it.

• 6

  Benchmarks should come from practitioners, not just labs: he says oddball tests like his own SkateBench and CM Griffin's Git bench matter because once a capability is measurable, researchers and model labs suddenly have a target to improve.