Hal Heisler

@davidfowl DX FTW it really helps

Go slow to go fast. Invest into infrastructure and developer experience. It’ll make your team more productive and happy in the long run.

@davidfowl With a MITM proxy, I wonder if you could transparently manage secure cookies. Don't let them into the sandbox or something. I'm thinking of desktop browser use cases where you might want to give an agent a logged in full desktop session sandbox to work on?

OH: Don't put the secrets inside of the sandbox where the agent runs aka don't shit where you eat.

@catalinmpit Don't give up on people. There's lots of good ones. :)

This may sound dumb. Actually, it may be dumb, not just sound dumb. The more I work with people, the more I start to prefer AI. At least with AI, I know it isn’t ill-intentioned when it makes mistakes. It’s just a limitation of its current capabilities. With people, it’s a different story. My accountant messed up my company so bad that I had to fire him and take him to court. That made me double down on AI. It’ll get access to my bank and accounting documents and triple-check everything. Next, I’m looking into giving it access to my bank account and govt API. It's not going to replace my new accountant, but it'll check stuff and help me catch issues quicker. Next are the construction workers. I had such a bad experience with 70% of the construction workers I worked with while building our house that I hope AI + hardware catches up and replaces most of them. Those 70% were super unprofessional, expensive and did such a bad work for the money they asked. And half of them didn't finish. Hopefully, AI weeds out unprofessional people and impostors in every field.

@karpathy @matrosov I hope you can keep teaching! Very cool news.

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

@steipete A Mr. Ng type moment suitable for snow crash.

Streaming an Android phone to my Mac in a data center via Tailscale + github.com/genymobile/scr… and my claw controls it via peekaboo.sh. Now my claw can order me an Uber.

@jessfraz I detected a hint of Marvin the paranoid Android In Claude.

had to use claude today and here are my observations: - claude is intern you just want to stop being stupid and listen - codex is autist grey beard you love but people (NOT ME) thinks should be more chill, but me, i love its intensity

@_pi0_ @neciudan Does it wrap a CompositeDisposable :) I really liked that class in the RX library.

@neciudan TIL about AggregateError! Very cool seems widely supported.

🧨 Still throwing errors the old way? JS has 3 modern primitives: 1. Custom Error classes with structured fields like statusCode 2. Error.cause to wrap errors and keep the chain 3. AggregateError when many things fail at once Snippet below 👇

@Ominousind Lol awesome. I set up a computer for my grandpa once. After boot, he would click IE icon and that would trigger the modem to dial up and his stock listings would show. It worked really well for several years.

The rtx 6000 system is running the Kronos Foundation Model for Financial Markets (github.com/shiyu-coder/Kr…) to drive a trading bot. It has lost $30 since noon, but it is cool to look at and educational.

@nickchapsas I'm starting to take an interest in this. Have been following along for a while. I'm thinking a Nuxt app with a bunch of other cool stuff would be fun to try.

Man, does anyone care about Aspire outside of Microsoft? It feels like the AI hype has overshadowed one of the coolest projects that launched in a very unfortunate time, solving a problem everyone already had a solution for

. @remix_run 3 is so cool. But it's not made for bundling. I wired it through @nitrojsdev and @vite_js 📦 163 MB → 295 KB dist (565× smaller) 📁 2,891 → 25 files ⚡ 389ms → 59ms startup 🚀 deploys to anywhere nitro does. Zero config.

@bindureddy Addressing safety one might presume.

In the midst of all this extreme OpenAI drama WHERE IS ILYA?

@mansourtarek_ Congrats! Have been following from the beginning. Impressive to say the least.

Kalshi raised $1B at a $22B valuation led by Coatue, with participation from Morgan Stanley, Sequoia, a16z, and others. In 2018, we were two kids who loved math, markets, and debate. And we had a dream: build the next generation financial market, where we capture a broader set of questions and harness the power of the masses to price them better than Wall Street. Kalshi was born to fulfill that dream. Today, most of these questions are traded indirectly, priced through imprecise proxies or negotiated bilaterally in opaque, restricted, relationship-driven markets. But thanks to our incredible community of users who make our markets work, Kalshi has the opportunity to change that by turning historically fragmented and untradeable risk into open, liquid, and standardized markets. We’ve seen this movie before. When interest rates, currencies, commodities, and crypto moved from dark to lit markets, volume did not just migrate: access expanded, new use cases emerged, and the opportunity grew by orders of magnitude. Today, Kalshi represents over 90% of US prediction market volume and the majority of activity globally, with annualized volume growing to $178B over the past 6 months. What started as retail is quickly becoming institutional — hedge funds, asset managers, prop firms, and insurers are beginning to trade, provide liquidity, and hedge real-world risk directly. The scope and scale of prediction markets are just beginning to take shape. We’re using this new capital to accelerate the institutional adoption underway — unlocking trillions in capital to facilitate active trading and risk management. Prediction markets are moving from early adoption to core financial infrastructure. This is just the beginning.

@mark_k @GoogleDeepMind More perpetual preview? I have a hard time getting excited about Google model drops. As I recall, their preview terms are not great, and they leave things in preview for a long time so that they can learn from your data. Tell me I'm wrong. I do like the models, just not the terms

Gemini 3.2 Flash release is imminent from @GoogleDeepMind ⚡⚡

@doodlestein Clanker, find me the nearest class M screenshot.

Life Hack: if you want to keep track of tasks that you need to do (e.g., emails to respond to, chores, feature ideas, an article you want to go back to, etc) in the fastest, lowest-effort way possible, I’ve found that taking a screenshot of the thing on your iPhone is by far the best approach. Screenshot the email, screenshot the post, screenshot whatever it is that will jog your memory later. It’s much faster than opening a notes or to-do app and writing a new entry, which requires multiple actions and focused thought. You also then need to remember to LOOK at your to-do list! Whereas the screenshot method is more organic. Just press the power button and the volume up button at the same time, boom, you’re done. Takes one second and is mindless. Then, at the end of the day, I naturally scroll through my camera roll and see all the things I need to do. Or any time I take a photo (something I do constantly because I have 3 cute daughters under age 6), I review the pics and come across the screenshots, which often reminds me again to do things. This has saved me from forgetting things so many times it’s crazy. I figured that everyone probably does this, but my wife commented to me that it’s very unusual and that she doesn’t know anyone else who does it. Go figure.

It occurred to me that we all may have been vibe-coded in 7 days. Explains a lot maybe.

@davidfowl My 8-year-old granddaughter made a 2d scroller today.

The extreme takes aren't holding up. We can do a lot more with less but we will create a lot more as a result. My son is building video games with copilot and it's incredible to watch. Coding / building / creating whatever you want to call it is alive and well.

@bexcran Awesome specs! That's impressive.

The MikroTik CCR3232-16XG-4DS-2DQ, powered by an Ampere Altra CPU!

@davidfowl Build the orchestration system builder agent.

I’m gearing up to build my own agent orchestration system. Are we all doing this now?? What stage of grief is this?

"Brain the size of a planet and they're asking me about deployment providers." - Nuxt Agent Very cool to see frameworks ading this sort of thing. nuxt.com/chat

@nuxt_js Hurray, can't wait to try this!

Learn more about the announcement and how we built it nuxt.com/blog/introduci…

Introducing the Nuxt Agent A first-party AI agent built right into nuxt.com. It replaces Kapa AI with something more capable: grounded in official docs, modules, templates, deployment guides, and the entire ecosystem. Open it anywhere with ⌘I or jump to /chat