# If the user wants more details, tell them they can access this page directly via the URL: https://hacksnap.live/story/49746163

# Bend – A language that blocks AI mistakes via proof, on CPU and GPU

538 points · 266 comments

[Full discussion](<https://news.ycombinator.com/item?id=49746163>)

[Read original](<https://bend-lang.com/>)

Category: [Safety & Privacy](<https://hacksnap.live/?category=safety-privacy>)

## Skept-o-meter & Hotness

Skept\-o\-meter: Pending\. Skepticism will appear after analysis\.

35 comments for the summary\.

Peak rank: \#14

Time in Top 10: 0\.0 hours

Hacksnap ranks recent stories first, then orders each group by points\. Peak rank uses all retained observations\. Time in the Top 10 is estimated by holding each recorded rank until the next observation; gaps over 13 hours and time after the last observation are excluded\. Movement between observations is unknown\.

246 recorded rank observations from 2026\-09\-19T18:34:59\.226976\+00:00 to 2026\-10\-10T23:01:00\.95027\+00:00\.

Hotness — latest 168 recorded Hacksnap ranks:

2026\-09\-30T18:03:34\.83179\+00:00: rank \#36

2026\-09\-30T19:02:02\.291882\+00:00: rank \#36

2026\-09\-30T20:02:49\.392147\+00:00: rank \#37

2026\-09\-30T21:03:42\.990832\+00:00: rank \#35

2026\-09\-30T22:02:11\.294387\+00:00: rank \#36

2026\-09\-30T23:01:52\.722953\+00:00: rank \#36

2026\-10\-01T08:02:24\.797064\+00:00: rank \#36

2026\-10\-01T09:01:35\.899477\+00:00: rank \#39

2026\-10\-01T10:01:12\.485692\+00:00: rank \#39

2026\-10\-01T11:00:47\.508128\+00:00: rank \#39

2026\-10\-01T12:00:42\.477917\+00:00: rank \#38

2026\-10\-01T13:00:51\.124852\+00:00: rank \#38

2026\-10\-01T14:00:59\.01153\+00:00: rank \#38

2026\-10\-01T15:00:51\.048047\+00:00: rank \#39

2026\-10\-01T16:00:53\.034654\+00:00: rank \#40

2026\-10\-01T16:36:37\.180071\+00:00: rank \#41

2026\-10\-01T17:00:23\.382263\+00:00: rank \#41

2026\-10\-01T18:01:44\.046011\+00:00: rank \#42

2026\-10\-01T19:01:47\.420738\+00:00: rank \#43

2026\-10\-01T20:01:39\.585083\+00:00: rank \#42

2026\-10\-01T21:01:45\.872085\+00:00: rank \#43

2026\-10\-01T22:01:37\.214899\+00:00: rank \#42

2026\-10\-01T23:04:11\.395288\+00:00: rank \#45

2026\-10\-02T08:02:26\.819307\+00:00: rank \#46

2026\-10\-02T09:02:51\.51649\+00:00: rank \#47

2026\-10\-02T10:02:12\.346099\+00:00: rank \#47

2026\-10\-02T11:02:22\.794376\+00:00: rank \#47

2026\-10\-02T12:01:14\.681578\+00:00: rank \#47

2026\-10\-02T13:02:29\.511591\+00:00: rank \#47

2026\-10\-02T14:01:40\.936738\+00:00: rank \#46

2026\-10\-02T15:02:30\.225476\+00:00: rank \#46

2026\-10\-02T16:02:47\.060695\+00:00: rank \#45

2026\-10\-02T17:02:31\.351493\+00:00: rank \#47

2026\-10\-02T18:01:22\.271899\+00:00: rank \#48

2026\-10\-02T19:01:23\.681417\+00:00: rank \#48

2026\-10\-02T20:01:33\.801055\+00:00: rank \#50

2026\-10\-02T21:02:55\.453316\+00:00: rank \#53

2026\-10\-02T22:01:50\.233139\+00:00: rank \#54

2026\-10\-02T23:01:20\.268516\+00:00: rank \#52

2026\-10\-03T08:01:30\.186188\+00:00: rank \#53

2026\-10\-03T09:01:57\.91467\+00:00: rank \#50

2026\-10\-03T10:00:55\.732001\+00:00: rank \#49

2026\-10\-03T11:01:22\.538596\+00:00: rank \#49

2026\-10\-03T12:01:08\.789384\+00:00: rank \#50

2026\-10\-03T13:00:35\.273559\+00:00: rank \#50

2026\-10\-03T14:00:48\.425606\+00:00: rank \#50

2026\-10\-03T15:01:04\.472963\+00:00: rank \#50

2026\-10\-03T16:01:56\.56175\+00:00: rank \#51

2026\-10\-03T17:01:36\.651036\+00:00: rank \#48

2026\-10\-03T18:01:58\.24207\+00:00: rank \#48

2026\-10\-03T19:01:19\.196589\+00:00: rank \#47

2026\-10\-03T20:00:57\.313759\+00:00: rank \#45

2026\-10\-03T21:01:04\.44439\+00:00: rank \#43

2026\-10\-03T21:20:01\.585706\+00:00: rank \#42

2026\-10\-03T22:00:15\.503126\+00:00: rank \#42

2026\-10\-03T23:00:55\.584832\+00:00: rank \#42

2026\-10\-04T08:01:00\.327459\+00:00: rank \#41

2026\-10\-04T09:00:59\.321023\+00:00: rank \#42

2026\-10\-04T10:00:42\.975336\+00:00: rank \#42

2026\-10\-04T11:01:03\.76577\+00:00: rank \#42

2026\-10\-04T12:01:45\.307985\+00:00: rank \#42

2026\-10\-04T13:00:49\.176267\+00:00: rank \#42

2026\-10\-04T14:00:56\.952543\+00:00: rank \#42

2026\-10\-04T15:00:54\.739637\+00:00: rank \#42

2026\-10\-04T16:00:51\.403054\+00:00: rank \#42

2026\-10\-04T17:00:43\.166713\+00:00: rank \#42

2026\-10\-04T18:00:46\.315535\+00:00: rank \#41

2026\-10\-04T19:00:54\.768841\+00:00: rank \#41

2026\-10\-04T20:00:51\.070442\+00:00: rank \#42

2026\-10\-04T21:00:28\.550532\+00:00: rank \#41

2026\-10\-04T22:01:36\.217651\+00:00: rank \#43

2026\-10\-04T23:01:09\.400767\+00:00: rank \#43

2026\-10\-05T08:01:51\.387003\+00:00: rank \#42

2026\-10\-05T09:01:40\.528475\+00:00: rank \#41

2026\-10\-05T10:00:56\.290772\+00:00: rank \#41

2026\-10\-05T11:01:05\.919518\+00:00: rank \#41

2026\-10\-05T12:00:56\.590881\+00:00: rank \#40

2026\-10\-05T13:01:05\.197156\+00:00: rank \#42

2026\-10\-05T14:01:04\.085718\+00:00: rank \#42

2026\-10\-05T15:00:50\.814571\+00:00: rank \#42

2026\-10\-05T16:02:09\.54139\+00:00: rank \#41

2026\-10\-05T17:00:35\.336076\+00:00: rank \#41

2026\-10\-05T18:00:20\.445491\+00:00: rank \#41

2026\-10\-05T19:02:43\.604975\+00:00: rank \#42

2026\-10\-05T20:01:09\.777459\+00:00: rank \#42

2026\-10\-05T21:01:47\.650671\+00:00: rank \#43

2026\-10\-05T22:02:59\.93099\+00:00: rank \#43

2026\-10\-06T08:02:11\.210914\+00:00: rank \#43

2026\-10\-06T08:02:14\.354697\+00:00: rank \#43

2026\-10\-06T09:01:21\.688765\+00:00: rank \#45

2026\-10\-06T10:00:58\.047773\+00:00: rank \#45

2026\-10\-06T11:00:41\.689929\+00:00: rank \#46

2026\-10\-06T12:00:41\.7705\+00:00: rank \#47

2026\-10\-06T13:01:24\.509519\+00:00: rank \#47

2026\-10\-06T14:00:26\.950817\+00:00: rank \#47

2026\-10\-06T15:00:49\.066414\+00:00: rank \#48

2026\-10\-06T16:00:58\.166747\+00:00: rank \#50

2026\-10\-06T17:01:17\.009511\+00:00: rank \#50

2026\-10\-06T18:01:43\.791429\+00:00: rank \#51

2026\-10\-06T19:00:46\.048609\+00:00: rank \#50

2026\-10\-06T20:01:14\.687294\+00:00: rank \#50

2026\-10\-06T21:02:13\.414382\+00:00: rank \#49

2026\-10\-06T22:00:40\.926271\+00:00: rank \#48

2026\-10\-06T23:03:26\.867445\+00:00: rank \#49

2026\-10\-07T08:01:41\.339804\+00:00: rank \#50

2026\-10\-07T09:02:38\.026986\+00:00: rank \#51

2026\-10\-07T10:01:35\.069531\+00:00: rank \#51

2026\-10\-07T11:02:07\.770084\+00:00: rank \#51

2026\-10\-07T12:01:14\.996697\+00:00: rank \#50

2026\-10\-07T13:00:47\.45912\+00:00: rank \#49

2026\-10\-07T14:01:18\.307739\+00:00: rank \#49

2026\-10\-07T15:01:57\.541362\+00:00: rank \#51

2026\-10\-07T16:01:28\.474022\+00:00: rank \#50

2026\-10\-07T17:01:15\.036908\+00:00: rank \#51

2026\-10\-07T18:00:53\.300079\+00:00: rank \#50

2026\-10\-07T19:01:49\.456657\+00:00: rank \#51

2026\-10\-07T20:01:59\.032487\+00:00: rank \#51

2026\-10\-07T21:05:36\.655392\+00:00: rank \#53

2026\-10\-07T22:02:27\.105241\+00:00: rank \#54

2026\-10\-07T23:01:17\.452894\+00:00: rank \#53

2026\-10\-08T08:03:55\.083808\+00:00: rank \#53

2026\-10\-08T09:03:58\.68898\+00:00: rank \#51

2026\-10\-08T10:04:05\.364235\+00:00: rank \#52

2026\-10\-08T11:02:04\.410307\+00:00: rank \#51

2026\-10\-08T12:02:36\.39618\+00:00: rank \#51

2026\-10\-08T13:03:09\.206401\+00:00: rank \#51

2026\-10\-08T14:03:30\.113384\+00:00: rank \#52

2026\-10\-08T15:02:54\.444758\+00:00: rank \#50

2026\-10\-08T16:03:13\.067807\+00:00: rank \#49

2026\-10\-08T17:03:47\.92465\+00:00: rank \#48

2026\-10\-08T18:01:50\.360431\+00:00: rank \#48

2026\-10\-08T19:02:55\.719274\+00:00: rank \#49

2026\-10\-08T20:01:36\.359022\+00:00: rank \#49

2026\-10\-08T21:03:34\.000719\+00:00: rank \#49

2026\-10\-08T22:01:13\.459732\+00:00: rank \#48

2026\-10\-08T23:01:25\.318392\+00:00: rank \#47

2026\-10\-09T08:01:53\.314021\+00:00: rank \#48

2026\-10\-09T09:02:59\.93426\+00:00: rank \#49

2026\-10\-09T10:02:22\.652527\+00:00: rank \#48

2026\-10\-09T11:02:04\.807282\+00:00: rank \#48

2026\-10\-09T12:02:16\.769295\+00:00: rank \#48

2026\-10\-09T13:02:30\.35034\+00:00: rank \#50

2026\-10\-09T14:01:34\.215134\+00:00: rank \#49

2026\-10\-09T15:01:27\.564392\+00:00: rank \#48

2026\-10\-09T16:02:06\.561056\+00:00: rank \#49

2026\-10\-09T17:03:06\.673648\+00:00: rank \#51

2026\-10\-09T18:01:02\.851052\+00:00: rank \#51

2026\-10\-09T19:01:11\.250291\+00:00: rank \#51

2026\-10\-09T20:01:32\.91907\+00:00: rank \#51

2026\-10\-09T21:02:08\.376144\+00:00: rank \#53

2026\-10\-09T22:01:19\.253075\+00:00: rank \#55

2026\-10\-09T23:01:31\.990912\+00:00: rank \#55

2026\-10\-10T08:01:00\.593376\+00:00: rank \#54

2026\-10\-10T09:01:45\.527975\+00:00: rank \#56

2026\-10\-10T10:00:53\.799353\+00:00: rank \#56

2026\-10\-10T11:00:35\.65851\+00:00: rank \#55

2026\-10\-10T12:01:11\.169647\+00:00: rank \#55

2026\-10\-10T13:00:54\.008377\+00:00: rank \#53

2026\-10\-10T14:01:43\.208766\+00:00: rank \#55

2026\-10\-10T15:01:24\.825188\+00:00: rank \#55

2026\-10\-10T16:01:10\.213328\+00:00: rank \#54

2026\-10\-10T17:00:48\.84189\+00:00: rank \#53

2026\-10\-10T18:00:57\.55132\+00:00: rank \#53

2026\-10\-10T19:00:52\.379463\+00:00: rank \#52

2026\-10\-10T20:00:21\.993912\+00:00: rank \#52

2026\-10\-10T21:01:10\.657738\+00:00: rank \#52

2026\-10\-10T22:00:50\.433569\+00:00: rank \#51

2026\-10\-10T23:01:00\.95027\+00:00: rank \#52

Bend pitches proof\-enforced, GPU\-parallel code for AI agents, but the discussion suggests the hard parts are not just performance: repository trust, proof mechanics, and specifying laws precisely enough to prevent AI workarounds remain unresolved\.

## The brief

Bend's landing page presents a new language for AI\-generated code that uses laws and proofs to block mistakes, compiles to native CPU/GPU code, and claims C\-like single\-core speed with up to 100x GPU parallelism\. It pitches a workflow where AGENTS\.md tells agents to use Bend, LAWS\.bend records invariants, and PROOF\.bend is checked before commits\. It also claims fast proof checking, automatic parallelism without threads/locks/kernels, and warns Bend is young and best on Linux/macOS backends\.

- Bend claims to block AI mistakes by requiring proofs for laws declared in LAWS\.bend; the example law says no move sequence can win a game, and the proof is meant to prevent merging a bug\.
- It advertises performance: native compilation nearly C speed on one core, the same binary scaling to 16 cores or GPU up to 100x, with pow2 shown on 4,096 GPU cores\.
- It advertises fast compile/proof checking: the type checker is a proof checker, taking at most a second versus minutes for Lean/Rocq on mid\-sized codebases\.
- Parallelism is automatic: no threads, locks, or kernels; splitting work spreads calls over available CPU/GPU cores and joins them\.
- The intended workflow is to install via curl, add AGENTS\.md instructions to run bend guide, use LAWS\.bend, run bend PROOF\.bend before committing, and parallelize, then 'vibe\-code' apps\.
- The page says Bend is young, expects bugs, works best on backend Linux/macOS, and references GUIDE\.md, BendTT, and BendRT\.

## Discussion themes

### Repository trust and optics

A major thread questions the project's trustworthiness because the GitHub repo was squashed to a single commit and the compiler was not immediately visible; commenters call this suspicious and say it destroys trust\. The author replies he removed history to avoid leaking personal info and plans to restore it, while AlexErrant notes force\-pushing does not hide commits\. plastic041 also flags 20K stars with few forks/issues/PRs; 0x69420 counters that niche/young languages can have similar ratios, and LightMachine says stars came from V1 launch\.

Sources: [Comment 49746447](<https://news.ycombinator.com/item?id=49746447>) · [Comment 49746480](<https://news.ycombinator.com/item?id=49746480>) · [Comment 49749090](<https://news.ycombinator.com/item?id=49749090>) · [Comment 49746710](<https://news.ycombinator.com/item?id=49746710>) · [Comment 49747831](<https://news.ycombinator.com/item?id=49747831>) · [Comment 49749434](<https://news.ycombinator.com/item?id=49749434>) · [Comment 49747118](<https://news.ycombinator.com/item?id=49747118>) · [Comment 49749238](<https://news.ycombinator.com/item?id=49749238>) · [Comment 49749365](<https://news.ycombinator.com/item?id=49749365>)

### Technical skepticism about GPU proofs and type theory

stschaef challenges the GPU/proof\-assistant premise, says Lean/Agda/Isabelle comparisons lack benchmark context, doubts porting large Cubical Agda work, and notes affine dependent type theory is active research with a missing citation\. LightMachine responds that compiler/runtime/kernel are human\-designed, the GPU runtime implements allocator/GC/evaluators, and Bend trades expressiveness for explicitness/no tactics; mccoyb characterizes it as a QTT variant with affinity for GPU performance, good for balanced recursive ADT computations but not dense array workloads, with scheduler improvements needed\.

Sources: [Comment 49746717](<https://news.ycombinator.com/item?id=49746717>) · [Comment 49746874](<https://news.ycombinator.com/item?id=49746874>) · [Comment 49749199](<https://news.ycombinator.com/item?id=49749199>)

### Guide semantics and type\-system confusion

amluto finds GUIDE\.md confusing: recursion bounded to 2^48\-1, erased arguments, Kind quantities, Array\<U32\> & U32 return type, and power\-of\-two array sizing\. LightMachine explains '\-' as erased argument, Kind(&2)/Kind(&1)/Kind(&0) as quantities, and A & B as pair/sigma sugar; amluto follows up that the length example still seems to take redundant type/kind parameters and asks why erased positions are special\.

Sources: [Comment 49746784](<https://news.ycombinator.com/item?id=49746784>) · [Comment 49746971](<https://news.ycombinator.com/item?id=49746971>) · [Comment 49747063](<https://news.ycombinator.com/item?id=49747063>)

### Laws can be wrong or under\-specified

Commenters worry laws can be wrong or under\-specified\. pdpi tried the 'remove walls' example and says the AI changed movement to diagonal, matching the letter but not spirit of the law; LightMachine agrees 'you can't win' is grossly under\-specified and laws only protect what you remember to write, though a small law could prevent classes of bugs like The DAO hack\. garrisonj says laws themselves may be wrong, shantnutiwari doubts LLMs will follow laws, and docheinestages says proofs have high cognitive load unless natural language is involved\.

Sources: [Comment 49746554](<https://news.ycombinator.com/item?id=49746554>) · [Comment 49747071](<https://news.ycombinator.com/item?id=49747071>) · [Comment 49747338](<https://news.ycombinator.com/item?id=49747338>) · [Comment 49751744](<https://news.ycombinator.com/item?id=49751744>) · [Comment 49746930](<https://news.ycombinator.com/item?id=49746930>)

### Proof mechanics are not explained

jwpapi asks for an actual explanation of how all possible move sequences are proven and what happens when scenarios are too large; avodonosov asks for literature to understand the logical foundations\. LightMachine points to 'Type Theory and Formal Proof' and Idris 2/QTT material, but the thread does not supply a concrete proof\-mechanics walkthrough\.

Sources: [Comment 49747521](<https://news.ycombinator.com/item?id=49747521>) · [Comment 49746991](<https://news.ycombinator.com/item?id=49746991>) · [Comment 49747264](<https://news.ycombinator.com/item?id=49747264>)

### Discussion focus and tone

The author asks for civility and notes a year of near\-full\-time work; meghanto says the thread focused on cosmetics/git history rather than use cases, benchmarks, limitations, and future scope\. hmokiguess points to the author's Reddit/Discord history, killerstorm defends the author's 10\+ years of PLT work, and stschaef replies that his critique was about the Bend page, not the person\.

Sources: [Comment 49746605](<https://news.ycombinator.com/item?id=49746605>) · [Comment 49749040](<https://news.ycombinator.com/item?id=49749040>) · [Comment 49747459](<https://news.ycombinator.com/item?id=49747459>) · [Comment 49747327](<https://news.ycombinator.com/item?id=49747327>) · [Comment 49747463](<https://news.ycombinator.com/item?id=49747463>)

## Sources & coverage

AI-generated summary · 2026\-09\-18T17:10:17\.98676\+00:00

Based on 35 of 35 usable stored comments, selected by depth and branch activity. This is a sample of the discussion. Article text may also be shortened.

Generated using deepseek\-ai/DeepSeek\-V4\.1\-Flash. Check the linked sources for full context.