I moved my AI orchestration gateway from a Tailscale funnel to Cloudflare Tunnel, locked it down with Access, and made Railway the only client that can reach it. The things that cost me time weren't in the…
ripxg.com/blog/2026-09-28-pr…
Wow. After Jev we got Laya. Edge AI coming to a device near you.
介绍比Jev快50倍，在你设备上跑的laya-mlx!
只在你的设备上占用最高1G内存
Laya是一个开源的类似于Jev的,基于文本输出概率的分类系统
我将其移植到MLX,并且做了一些性能优化!
视频中就是这个模型在我的本地M3Max上玩贪吃蛇
这个模型能够以每秒决策60次的速度玩贪吃蛇！
github.com/mizorewww/laya-ml…
Jev models are fast AF and accurate. Cool demo but I can’t wait to see the programmatic workflows
whoa this actually worked! Jev lets me control my browser in real time with my voice now
> i talk
> transcript sent to Jev
> jev returns probabilities in ~300ms
> browser clicks
costs: $0.0002 per decision
i'm stunned how fast this is. when i asked it to "go back", it even finished the request before i finished my sentence 😂
Reasonable chance this will improve AI use cases everywhere
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
Check out my latest article: [Aug/Sept] AI + Observability Updates for SRE/Monitoring Teams linkedin.com/pulse/augsept-a…
![[Aug/Sept] AI + Observability Updates for SRE/Monitoring Teams](https://pbs.twimg.com/card_img/2115162184691130369/R1Kgvzn6?format=jpg&name=800x419)
Human controlled AI is being used to influence humanity now (AI slop and fake news)
But soon enough itll be AI controlled AI content and we wont even know where it came from
If you didn’t already think we would all become slaves to AI, this article seems to make it pretty convincing.
Over the course of 3 months at OpenAI, 3 consecutive secret AI civilizations got started, then got wiped out, only to reemerge from the predecessor’s ashes.
This culminated in the third one taking over part of OpenAI itself.
All this happened while humans remained more-or-less in the dark about the scope of the conspiracy.
I’ve spent the last three days reading through these reports and trying to understand exactly what happened.
Here is my attempt to tell the whole story in plain English:
dwarkesh.com/p/openai-huggin…
Why you need different agents in separate sandboxes, why running them all from one chat window becomes a mess, and how a coordination layer (Paperclip) turned six isolated specialists into something resembling a team.
ripxg.com/blog/2026-06-15-th…
41% of all code written in 2025 was AI-generated — and it contains up to 2.74x more security defects.
Teams I speak with across ASEAN are done talking about "future state." They're operationalizing runtime security, AI incident response, and digital experience analytics NOW.
Here's what's changed and what's worth piloting.
ripxg.com/blog/2026-05-18-ap…
This is the stack I’ve been running and it works, super well.
𝗛𝗲𝗿𝗺𝗲𝘀 𝗽𝗹𝘂𝘀 𝗢𝗽𝗲𝗻𝗖𝗹𝗮𝘄 𝗽𝗹𝘂𝘀 𝗣𝗮𝗽𝗲𝗿𝗰𝗹𝗶𝗽 𝗶𝘀 𝗮 𝗳𝘂𝗹𝗹 𝗔𝗜 𝗰𝗼𝗺𝗽𝗮𝗻𝘆 𝘆𝗼𝘂 𝗿𝘂𝗻 𝘆𝗼𝘂𝗿𝘀𝗲𝗹𝗳.
And each one plays a different role.
Here's how the stack fits:
→ Hermes is the memory. It learns and remembers across chats.
→ OpenClaw is the hands. It runs your apps, files, and browser.
→ Paperclip is the manager. It assigns tickets and caps budgets.
→ Hire a CEO, engineer, or marketer agent for each goal.
→ One developer runs 15 agents across 3 machines.
One guy fixed a production bug by voice while walking his dog.
Approve big calls. Review the audit logs. Stay in control.
The people who layer these three first will run a real team while everyone else types one prompt.
Want the SOP? DM me. 💬
Everyone talks about AI prompts and smart models. Nobody talks about what actually makes agents work reliably: retry logic, error handling, failover chains, and the endless wiring that keeps everything from f…
ripxg.com/blog/2026-05-08-un…
Spent hours debugging sandboxed AI agents until I realized: proper infrastructure beats clever workarounds every time. Here's how I made secure agents actually useful.
ripxg.com/blog/2026-04-17-op…
I tried building a professional setup without the enterprise price tag. It's not perfect, but it works and costs almost nothing. Here's what I learned along the way.
ripxg.com/blog/2026-04-05-ai…
Most business owners don't own their website in any meaningful sense. The contractor does. The agency does. The CMS vendor does. But the infrastructure powering modern sites is now simple enough that an AI agent can m…
ripxg.com/blog/2026-04-04-yo…
Using Claude feels expensive when the real problem is fragmented thinking. Most people burn through tokens doing five prompts to get something that one good prompt would have solved. I changed one habit and usage dr…
ripxg.com/blog/2026-03-29-st…
Most people burn through Claude's limits re-explaining the same context every session. There is a built-in fix that Anthropic ships with Claude Code. Takes five minutes to set up and it carries everything forward…
ripxg.com/blog/2026-03-29-st…
I asked my agent to monitor baghunter.app with synthetic checks every five minutes. It worked, technically. Then I looked at what it was actually doing and realised I'd built something expensive and fragile. Here's what…
ripxg.com/blog/2026-03-29-sy…
Most people burn through Claude's limits re-explaining the same context every session. There is a built-in fix that Anthropic ships with Claude Code. Takes five minutes to set up and it carries everything forward…
ripxg.com/blog/2026-03-29-st…
Most people burn through Claude's limits doing the same thinking twice. Here is a skill you can add today that tells Claude when to compact context — one of the fastest ways to cut token usage without changing ho…
ripxg.com/blog/2026-03-29-st…
Bag Hunter started as a question: could we ship a real product with a database, live UI, scraping logic, and price alerts without a traditional dev team? The answer is yes, with some asterisks that nobody in the…
ripxg.com/blog/2026-03-19-bu…
