wrote about why “metaframework is dead” feels true rn.
not bc frameworks are useless, but bc @OpenAIDevs Codex + @claudeai Claude Code make explicit scaffolding cheap. i’d rather codegen real routes/contracts than hide more runtime magic.
https://t.co/15U9hDqf9h
A package is a distribution boundary.
Use package names for code that leaves the monorepo. Use root imports for internal source. Enforce architecture with visibility/lint rules, not fake package.json files everywhere.
https://t.co/2Nbdleocqn
Last part is up: a frontend build system is product infrastructure.
@bazelbuild gives you precision, but teams still need rule APIs, codegen, docs, good errors, metrics, and gradual adoption so the graph stays usable.
https://t.co/LG06YLpWFL
Part 7 is up: @bazelbuild, @turborepo , and @NxDevTools are not one ladder.
Turbo runs package tasks. Nx models projects. Bazel models actions and artifacts. The right choice depends on where your repo’s complexity already lives.
https://t.co/IFpGkwlIj7
Part 6 is up: a successful bundle is not the same as a correct artifact.
Use @bazelbuild to verify built outputs, then make deploys consume those verified artifacts instead of rediscovering files from the working tree.
https://t.co/ROyRZrECdY
Part 5 is up: @bazelbuild should not replace Vite.
Bazel decides what the app needs; Vite turns it into a bundle. Apps, SDKs, tests, Storybook, assets, and output checks all become explicit graph edges.
https://t.co/yaqDwHwiLN
Part 4 is up: some frontend inputs are graph questions.
Translations, icons, routes, and generated TS packages get cleaner when @bazelbuild can model them as named artifacts instead of side effects.
https://t.co/KYv1YQYl1t
Part 3 is up: typechecking is not one giant tsc --noEmit.
Split runtime, test, config, generated, and lint contracts so @bazelbuild can catch dependency leaks where they happen.
https://t.co/izDDcsuJbi
Part 2 is up: frontend packages are rarely just TypeScript folders.
This one breaks down runtime code, tests, stories, config, assets, visibility, and codegen as separate build surfaces so @bazelbuild can rebuild less and explain more.
https://t.co/U2dPeVhrLk
Part 1 of my series on @bazelbuild for large frontend monorepos.
At scale, the hard question is not “how do we build the app?” It is “what did this change affect?”
https://t.co/gCW3kCc3HV
Introducing Perplexity Computer.
Computer unifies every current AI capability into one system.
It can research, design, code, deploy, and manage any project end-to-end.