Been noticing how quickly vibe coding has become a real workflow lately.
A few months ago most of us were still writing everything manually or just using AI for small snippets. Now it feels like the process has shifted to describing what you want and iterating with the AI until the product behaves the way you imagine.
I've been experimenting with a workflow where I give an AI a predetermined test suite and ask it to write only enough code to pass those tests. Think of it as "building the app from the tests."
What's surprised me is how hard it is to stop the model from doing extra work. Smaller models especially seem to ignore constraints, they'll refactor, add features, or change things that weren't requested.
For those of you using Claude Code (or similar tools), how much do you rely on .claude files and hooks to enforce behavior versus just prompting the model?
If you don't use hooks, what's your approach for keeping the AI inside the boundaries you want? Do you have a workflow that's been reliable?
Every time I've tried one of these tools, a chunk of what I end up spending is just fixing something it broke, not adding anything new.
And it's not a fixed cut, it gets worse the more complex the app gets, simple screens barely need any fixing, anything with real logic starts eating credits fast.
Obviously there is a ton of hype around OpenClaw and everyone rushing to get Mac Minis and set it up. But what happens now that Anthropic has created a seemingly more secure and easier to set up version of OpenClaw with Channels? Do you think OpenAI will adapt quick enough and evolve OpenClaw to be at the forefront of the localized agent space? Or will Anthropic run away with it? Would love to hear what you all think. Btw, our current operation at Honestly is using Claude Code Channels, check out our socials accessed through our #4 PH rank to follow along with how we've been using them!
Since agents will happily rebuild anything you ask in minutes, "I'm afraid to touch this" stops meaning what it used to. But I don't think the fear goes away, it just moves.
I keep a section in my CLAUDE.md that isn't architecture notes, it's just "already tried this, here's why it didn't work." Without it the agent will quietly rebuild the same mistake in 3 months, because it has no memory of the first time and the code looks perfectly reasonable on its own.
sharing something I built that might be useful here.
It's an open-source CLI tool called AI SOV Analyzer. The idea: AI models like ChatGPT are increasingly where buyers discover products. This tool lets you measure how often your brand shows up in AI responses vs your competitors basically "Share of Voice" but for AI.
Free to run (works with Ollama locally or free-tier APIs)
