Aver – a language designed for AI to write and humans to review
I’ve been building Aver around a simple question: If AI is going to write more of the first draft, what should the source look like for the human reviewer? Aver is an experimental statically typed language and toolchain for AI-written, human-reviewed code. My bet is that source code should carry more than implementation. In most projects, implementation survives in the code, but intent lives in docs, decisions live in ADRs or tickets, and expected behavior lives in tests that may or may not stay close to what they describe. Aver tries to make those parts first-class: explicit method-level…
What it does
In the maker’s words, at launch
I’ve been building Aver around a simple question: If AI is going to write more of the first draft, what should the source look like for the human reviewer? Aver is an experimental statically typed language and toolchain for AI-written, human-reviewed code. My bet is that source code should carry more than implementation. In most projects, implementation survives in the code, but intent lives in docs, decisions live in ADRs or tickets, and expected behavior lives in tests that may or may not stay close to what they describe. Aver tries to make those parts first-class: explicit method-level effects in function signatures; ? strings for machine-readable function intent; decision blocks for design choices and tradeoffs; colocated verify blocks for pure functions; deterministic record/replay for effectful flows; aver context for compact contract-level module export; aver compile to Rust; and aver proof to Lean 4 for mechanical proof checking of the pure subset. A small pure example: fn charge(account: String, amount: Int) -> Result<String, String> ? "Pure charge validation and transaction-id creation." match amount 0 -> Result.Err("Cannot charge zero") _ -> Result.Ok("txn-{account}-{amount}") verify charge charge("alice", 100) => Result.Ok("txn-alice-100") charge("bob", 0) => Result.Err("Cannot charge zero") A reviewer can see at a glance what the function is for (`?`) and a machine-checkable example of expected behavior (`verify`). An effectful wrapper looks like this: fn chargeAndPrint(account: String, amount: Int) -> Result<Unit, String> ? "Effectful wrapper around charge." "Prints the transaction id on success." ! [Console.print] result = charge(account, amount) match result Result.Ok(txn) -> Console.print(txn) Result.Ok(Unit) Result.Err(err) -> Result.Err(err) The `!` makes side effects part of the signature rather than hidden inside the implementation. Aver is intentionally opinionated: no exceptions, no null, no `if`/`else`, no loops, no closures. Branching goes through `match`, failure through `Result`, absence through `Option`, and side effects are explicit. The repo includes small examples, but also `projects/workflow_engine`, which is my attempt at a medium-sized auditable application core with app/domain/infra split, replayable effectful flows, and verify-driven pure logic. This is still early. I’m not claiming everyone should replace mainstream languages with Aver. The narrower question I’m testing is whether making intent, decisions, checks, and effect boundaries machine-readable inside the source makes AI-produced code easier to review, constrain, and trust. I’d especially like feedback on whether this feels like a language worth existing, or whether the same idea should just be conventions and tooling on top of an existing language.
Does the same job
all alternatives →
- WBWe built an AI to review your pull requests2025 · infinitcode.ai · ▲53
Hey HN, We’re two developers (co-founders) with a team of 20 who got tired of spending hours reviewing PRs, so we built Infinitcode.ai, an AI-powered code reviewer that: - *Summarizes PRs in plain English*: No more deciphering 1,000-line diff jungles - *Catches more than bugs*: Security holes, performance pitfalls, code smells, even typos (yes, we’ll flag “vurnerabilities” and vulnerabilities) - *Zero onboarding*: Works instantly—no “let me learn your codebase for weeks” nonsense. Why we’re posting: We’re in alpha and need brutal honesty. Roast our tool, mock our UI, or tell us why AI will…
- QAQuibbler – A critic for your coding agent that learns what you wantOct 2025 · github.com · ▲114
- RARevise – An AI Editor for DocumentsMar 2026 · revise.io · ▲85
I started building this 10 months ago, largely using agentic coding tools. I've stayed very involved in the code base and architecture, and have never moved faster in my life as a dev. The word processor engine and rendering layer are all built from scratch - the only 3rd party library I used was the excellent Y.js for the CRDT stack. Would love some feedback!
- DADrafting AI – Human-in-the-loop AI automation for ops teams' email2024 · getdrafting.com · ▲29
Hi HN, I'm Jared from Drafting AI (https://getdrafting.com/). We're a Chrome extension that helps operations and customer support teams work through their inboxes more efficiently. Here's a quick demo video: https://www.youtube.com/watch?v=PCDqAMaYx2Q We believe AI can be deployed more widely when you mitigate hallucinations through human-in-the-loop review. This led to two decisions: 1. A Chrome Extension is ideal for presenting actions in third-party apps for human review. When AI suggests issuing a refund via Shopify, what better way to show it than through…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
Launched alongside, March 2026
the whole month →
Switch from ChatGPT to Claude with import memory feature
AI · Mar 2026 · claude.com
