You know what’s more sobering? Amazon *already* has some of the most stringent/sophisticated CI/CD game in the business. See https://t.co/D7Upm1VTZD Most embracing “vibe coding” don’t have the guardrails Amazon already does and are going to relearn reliability first principles.

You know what’s more sobering? Amazon *already* has some of the most stringent/sophisticated CI/CD game in the business. See https://t.co/D7Upm1VTZD Most embracing “vibe coding” don’t have the guardrails Amazon already does and are going to relearn reliability first principles.

@copyconstruct really good point

@Strakyo Yes

@pvncher @RepoPrompt Congrats and happy birthday to RepoPrompt. So many of my friends use it to code.

@AlexFinn @moltbook It's the only one I see discussed on X. You got another that's better?

HF infra announces their new tech and a plan to revolutionize cloud storage Introducing Storage Buckets: storage made more accessible and more versatile Using Xet for deduplication and with a CDN, here is how it works + why it has nothing to do with traditional cloud storage🧵 https://t.co/3d0GjGU6AS

Introducing Upstash Box - the best way to give your AI agents a computer 🎉 ◆ Secure, isolated cloud sandboxes ◆ Built-in Claude Code, Codex or OpenCode ◆ Sleeps when idle, wakes up in milliseconds https://t.co/LP2WI8A475

GPU availability continues to plummet. New multi-year lows in availability for: B200, H100, and A100s. Good luck getting a B200 (availability <5%). Oil getting all the headlines (rightly), but, for markets, the dominant secular story is surging compute demand. https://t.co/OcyJYHQpXp

Last week I shared that Truesight now works anywhere via MCP. Here's what it looks like in practice. Image 5 is the one worth looking at first. Three customer support traces, two binary evals each. Two pass. One fails, and Truesight tells you exactly why: the AI agent told a customer with a missing package to file a claim themselves instead of initiating a carrier investigation. That's not a 0.34 helpfulness score. That's something you can actually fix, grounded in your company's policies. The rest of the images show how you get there. You describe what you want to quality control in plain language inside your editor. Truesight asks a few scoping questions, proposes a quality control design with explicit pass/fail criteria, and deploys live API endpoints when you approve. Image 3 shows why it recommends two separate binary evals instead of one combined score: if accuracy and empathy collapse into a single result, you lose the signal about what actually broke. Image 4 shows how it auto-generated a companion skill so you can run the same evals against new traces anytime without rebuilding anything. The whole thing took about 6 minutes and never left the chat. If you want to catch that kind of AI failure before your customers do, sign up at https://t.co/bwdXrVMK60 or DM me directly.

+3 more

@amankhan @swyx I’ll save one for you 🙌

very proud to have an actual prop to wear when i ask ppl to “put their evals hat on”

jd vance, at all times, in front of every audience, under any and all circumstances: https://t.co/5ltTKruFsM

reconnecting 5/4

Have you tried forking chat sessions in VS Code? Create a new independent session that keeps the conversation history so you can explore alternative approaches without losing your original context. Watch the video to see how it works. https://t.co/YZRm5f58HS https://t.co/THuAWN0PMF

@rkienle13 did you get stabbed with a fork?

@mitchellh it's fun. i have built my own to experiment with new ideas like proactive agents, evals, and figuring out interesting ways to interact with ai agents. it has now turned into an orchestrator app that takes up 95% of my time. other 5% is tuis

i will be posting more detailed thoughts on this in the coming days, and how to experiment around agent harnesses. stay tuned.

This has been our fastest growing recent product. AI WANTS data. We’re making petabyte storage cheap and fast.

Extremely bullish on the future of AI native wearables. lots of new form factors and capabilities open up with voice and video. There's at least one surface that's deeply under-explored today, but ultimately I think consumers under 50 will be wearing an average of 3 computers in the next 3-5 years to augment smartphones.

Yann just bet a billion dollars that the entire industry is building on the wrong foundation. Large language models predict the next word. They're trained on text, so they understand language. But the real world isn't made of words. It's made of continuous sensor data: camera feeds, touch, sound. And most of that data is unpredictable. You can't predict every pixel in a video the way you predict the next token in a sentence. Generative models fail here because they try to predict everything, including noise. AMI Labs is building world models using JEPA (a method LeCun proposed in 2022 that learns abstract representations of reality and predicts in that compressed space, not in raw pixels). Action-conditioned versions let AI simulate the consequences of actions before taking them. That's not generation. That's understanding. This unlocks AI that can operate in the physical world without hallucinating: 1. Robotics that plans multi-step actions 2. Healthcare devices where errors kill patients 3. Industrial process control under safety constraints 4. Wearables that adapt to real-time sensor input If JEPA works at scale, the next wave of AI companies won't fine-tune LLMs. They'll train world models on sensor data. LeCun's CEO already predicts every startup will rebrand as a "world model company" within six months. The architecture war is starting.

Here’s what’s gonna happen: - you replace your code review with feedback loops (sentry, datadog, support tickets, etc) - you stop reading the code - software factory fixes everything - one day something breaks at 3am, agent can’t fix it - nobody’s read the code in 3 months - you have 3 weeks of downtime trying to re-onboard and fix it - you lose significant % of your contracts and users - your company is now dead

Here’s what’s gonna happen: - you replace your code review with feedback loops (sentry, datadog, support tickets, etc) - you stop reading the code - software factory fixes everything - one day something breaks at 3am, agent can’t fix it - nobody’s read the code in 3 months - you have 3 weeks of downtime trying to re-onboard and fix it - you lose significant % of your contracts and users - your company is now dead

this is so sf i love it 🤣 @frontiertower has a vending machine run by an @openclaw agent selling @bryan_johnson 's @bp_blueprint longevity products https://t.co/u3aExZKKm4