WebDevCody

@shariqriazzz @cursor_ai I'm not sure honestly, here is my export from cursor usage

@webdevcody @cursor_ai wtf is xhigh? grok 4.5 has high reasoning top via api and grok build lol.

Just crossed 1 billion total Grok 4.5 tokens using @cursor_ai

@aipulseda1ly @cursor_ai no one is on your level

@webdevcody @cursor_ai try 1 billion in fable in one day

@ziadeh50 yeah, just a little joke, I appreciate your contributions!

@webdevcody I wouldn't really call that a hack, people contribute because they believe in the project and want to help make it better

coding hack, make a project open source during the Fable 5 release and get strangers to use their claude subs to just add a bunch of features.

@_hakagie @cursor_ai it's def not as smart as the other frontier, but I'm also not kicking off multi day plans. Often I just need a quick feature addition or bug fix, and I'll iterate on that until it's good. I've found regardless of the model, you have to always reprompt to fix gaps and edge cases.

@webdevcody @cursor_ai did they make an actually good model?

@orcdev if VideoOrc let users edit and record videos efficiently and easily, I'd say your ai generated code is working just fine

if you're generating millions of lines of code a day, there's only one way to know your AI generated code is actually working without reading every line

We live in the era of the great humbling, where skills you spent years learning can be outperformed by a single prompt. You may not be as important as you thought you were. Now go build your self-worth around something technology can’t easily commoditize.

A senior engineer’s job is to evaluate new technology and determine whether it improves the product, the team, or productivity. If your response to AI is blind hostility instead of serious evaluation, you’re not a senior. AI has great use cases and pitfalls. Learn both.

@Rasmic hate the game not the player

When I tell you videos like this piss me off First, most of these people make money through selling you some garbage make money online course… you learn nothing of value… they sell you a dream and you buying makes them richer Second, it’s videos like this… I’m so rich idk what to do with my money… these videos do extremely well with someone who has nothing and these end being the people that buy As someone who had nothing and came from nothing… these videos/people piss me off

@TheCoderBtw either way, the longer the context window, the more the model knows not to re-review the same stuff, but the orchestrator could be smart enough to kick off a bunch of individual isolated reviews.

@webdevcody Is this useful in the same thread? I think this would be especially useful if a separate agent is reviewing the work, so the context doesn't get shared.

the key to agentic coding

@thingsdev14 yes, I use a workflow which will kick in various sub agents based on the type of changes in the branch. If lots of api endpoints were changed, a security audit is invoked, etc, but even with that, it still misses a lot of stuff and you have to re-review constantly.

@webdevcody I'm new to defining AI workflows but do you think there's value in specifying what each of those passes on the codebase could be checking for?

@nick_woodward_ @dreamsofcode_io no, I'm not. show me code that handles all potential status code errors from a third party service vs one that doesn't. guess what, the robust code that handles all failure scenarios and properly retries, logs, rollback transactions will have more lines of code

@dreamsofcode_io @webdevcody He knows. He's engagement farming.

Why is there so much anti-LoC rhetoric? At some point, more features, edge cases, security, and reliability require more code. There’s no magic abstraction that lets you build an operating system in 500 lines. Unless you’re building a todo app, you’ll need lots of code.

@theognisSC2 been using cursor cli a lot, but I use goal as well

@webdevcody curious, why do you use /loop and not /goal?

the biggest issue I have with using ai are the gaps and drift they leave when you make changes. a simple "/loop remove as much dead code from my project until you get it all" will show you that all the previous prompts you made with AI left a bunch of stuff.

@kevin_moechel Static analysis finds predefined patterns. AI can reason about intent and context. Will fallow audit find abandoned Terraform resources, overly broad IAM policies, obsolete feature flags, or code that still runs but no longer serves a purpose? They’re complementary tools.

@webdevcody pnpm dlx fallow audit Why trust non-determinism to find something that's perfectly fine too evaluate with static analysis? And all the tokens wasted..

@ahmed_eumar the P0-P1 I wouldn't ignore, especially if using frontier model to do the security audit

@webdevcody Most of my experience with this is one or two geniune issues and the rest are non-issues, false positives ,etc... Is this yor experience as well?

"/loop continue to explore my codebase using sub agents for security issues and create a github issue for each one with the proper P0-P4 label"

@alonshrestha did I say success anywhere in my post?

@webdevcody so? you mean code volume equal success. i'm more worried if you used ai to build this content.

just remember, some kid in college with multiple Claude Code subscriptions using Fable 5 is producing more code in the past few weeks than you ever have in your entire 10+ software career.

@thingsdev14 yes, the model looks into something new to investigate each prompt from what I'm seeing

@webdevcody Is there any difference between the checks?

"Hey, so welcome to the interview! before we start I just want to make sure you're ok that we use Gemini to record our conversation so that we can feed it to an LLM to determine if you're a good culture fit or not. Ok? First question, What’s your biggest weakness?"

@mattpocockuk then yes, make more roadsigns for ai

@webdevcody I consider those roadsigns rather than docs

Docs explaining the code is wild, I've never done that. High level system architecture docs are useful though otherwise you'll just watch these agents re-explore the same files trying to "understand" how it works constantly (thus using more tokens each new session).