hraness

saved

Bend

by BendBend

Hraness cites a source capture. The source author remains the source.

gist

Bend’s public site pitches a programming language for post-AGI vibe coding: near-C single-core speed, automatic multi-core and CUDA parallelism, and Lean-style proofs that run fast enough for an agent to check after every change. Intent is declared as laws in LAWS.bend; PROOF.bend must show they hold before commit, so AI mistakes that break those laws cannot merge. The landing page is product prose with install steps, AGENTS.md instructions, a guarded-game demo, and links to GUIDE.md plus BendTT/BendRT papers—not a long essay.

ideas

  • Laws beat ambiguous prompts. Humans stop reading code but still need precise intent; laws and proofs replace fuzzy natural-language acceptance for what must never break.
  • Speed is agent-shaped. Native compile near C, automatic parallel/GPU scale, and sub-second proof checking so agents can verify after each edit instead of waiting on slow dependent-type tooling.
  • LAWS.bend is AGENTS.md with teeth. Declared laws make “make no mistakes” type-checked; without them a wrap-around board change can merge a broken invariant, with them the agent must repair and prove before ship.
  • Young backend-first stack. Install via curl script, point agents at bend guide, LAWS, and PROOF, prefer Linux and macOS backends, and expect bugs while the language evolves.

quotes

a fast language that blocks AI mistakes via proof

Bend, stating the product thesis.

How can you trust code you never read? By demanding a proof.

Bend, on unread AI-written code.

Merging a bug is mathematically impossible: it is a theorem.

Bend, describing law-guarded merges.

LAWS.bend is AGENTS.md backed by proof.

Bend, equating laws files to agent instructions.