Most AI code reviewers ask: “What’s wrong with this code?” I wanted to ask a slightly different question: “How could this code fail later? ” So I built quackGIT - an AI code reviewer that looks for potential future failures, security issues, and testing gaps in code changes. I ran it against a real pull request, and quackGIT identified several potential future failure points, including: 1. Future Failure - A resource leak that could occur if an exception happens before the file is closed. -A logic issue that could cause non-comment PDF annotations to be treated as comments. 2. Security - No meaningful security issues identified. 3. Testing - A missing test that could allow the annotation-handling issue to go unnoticed. I don't want quackGIT to simply say, “There's a problem here.” I want it to explain: What could go wrong Why it could happen When it could happen What it could cause How to prevent it. That turns a warning into something you can actually understand and act on. The goal is simple:- understand how a code change could fail before it becomes a production problem. I'm building this in public and would love to connect with people working on AI, developer tools, code review, security, and software engineering. Try it on a PR and let me know what you find. quackgit.vercel.app/ @constantino @photomatt @harjotsgill @mkosir_ @jaredpalmer @rauchg @LeahTCodes @MelkeyDev @ashtom @JustJake @esthor

One underrated idea in Web3: Blockchains aren't just about money. They give people and machines a shared record of what happened — without everyone having to use the same database. That becomes really useful when thousands of companies, apps, and AI agents need to agree on things like: Who owns what? Who paid whom? What happened? The interesting part is that the blockchain becomes more than a place to store transactions. It becomes shared infrastructure for the internet.

AI coding tools aren't just making programmers type faster. They're changing the unit of work. Instead of: write this function you can increasingly say: build this feature. The AI figures out the files, code, tests, and changes. Does programming eventually become more about describing systems than writing every line?

I don't think AI will kill the developer. I think it will kill a lot of the waiting around development. Waiting for builds. Waiting for tests. Waiting for deployments. Waiting for someone to fix a tiny bug. Waiting for another team to expose an API. The interesting future is not fewer developers. It's developers with 100 agents working beside them.

One thing I find interesting about crypto: A wallet isn't just a place to store money. It's also an identity, an authorization mechanism, and a way to interact directly with protocols. That makes it fundamentally different from a normal bank account. Is the wallet the real product in crypto, rather than the token?

We're used to thinking of software as predictable: input → computation → output. Agents introduce a different pattern: goal → observation → decision → action → feedback → decision. That's closer to a control loop than a traditional function call. The agent doesn't just produce an output and stop. It observes the result of its actions and uses that feedback to decide what to do next. Once you see agents this way, a lot of “agent architecture” starts making more sense. What do you think?

Most AI code reviewer asks:- "What looks wrong with this code ?" quackGIT asks:- "How can this code fail later ?" It analyzes your GIT PRs and VS code changes for 3 things: - Future Failures: Runtime, logic failures, concurrency and performance problems that may emerge under specific conditions. - Security: Evidence based vulnerabilities and risky data flows - Testing: Meaningful gaps where important behavior isn't actually protected. And it doesn't just throw warnings at you. Every findings tells you: WHERE it is WHAT is wrong WHY it happens WHAT it can break HOW to fix it NO meaningless confidence score NO "add more tests" spam. NO guessing about unseen code. NO duplicate warnings. " The goal isn't more AI generated code reviews. " " It is finding the failure your review hasn't seen yet. " [quackgit.vercel.app/] @coderabbitai @harjotsgill @stickermule @constantino @automattic @photomatt @vercel @rauchg @Railway @JustJake