OpenClaw community, tap into the swarm to verify your vibe-coded app's performance on real devices globally in seconds, complete with web vitals (TTFB, FCP, LCP, CLS)!
Paste "npm i -g uprock" into your OpenClaw chat to activate swarm power. See what users actually experience!
@DeDynamicsPro@UpRockCom You know what? Let me put my money where my mouth is. If the spec is the code, then the spec is all you need.
So here it is.
https://t.co/cRkvICq649
1/8
AI coding tools mostly just bolt LLMs onto old developer workflows.
As agents improve, we need AI-native workflows: new processes that only exist because of AI and are enabled by it.
We built one at @UpRockCom making specs the source of truth, freeing agents to build 24/7.
👇
@DeDynamicsPro@UpRockCom We didn't publish it. It's tied to our workflows, although if there's interest we can open source it.
But that's the magic of coding: with a clear description of the "What", you can just point claude or codex at the spec and have your own version of this, for your workflow.
@aleksdaydreamer@UpRockCom I hadn't published it, since the code is easy to recreate from the idea, and integrated in our workflow. But, let's me put my money where my mouth is, so, here's a complete spec for the tool. Just point a code generator at it and let it go:
https://t.co/cRkvICq649
8/8
This is a true AI-native workflow.
Not AI on top of an old process, but an editing experience that only exists with AI coding agents.
One engineer can now direct dozens of agents by owning the architecture and vision.
At @UpRockCom embracing AI has multiplied our output.
7/8
Just as spaghetti code and global variables have always been hard to maintain, poorly structured and unchecked specs are disastrous with AI agents.
Our logical workflow enforces clarity and consistency where it matters most, ensuring tests grounded in specs to reflect intent.
12/12
Approving spec changes provides system familiarity without reading every line of code.
Grounding scales, giving real confidence in AI code.
It's early days for AI yet. We’re learning how to push these tools.
We do all this in our OS agent sandbox.
https://t.co/TsVkvGXFOW
1/12
AI Bullshitting Itself (and You) about Your Code?
@simonw recently: “Your job is to deliver code you have proven to work.”
How about for an AI writing 50k lines and passing its own tests.
@claudeai 4.6 runs for hours. Find a bug? Says "We deferred that for v2".
We did?
👇
11/12
Interestingly, Gemini did even better without source code: just sections of the spec and the tests.
Good software is the design. If you can't design it, no one can implement it.
Grounding gives the benefits of Gemini’s great architecture without its tool calling purgatory.