Gaute Løken

@mattpocockuk The same goes for coding, honestly. Ask it to investigate party of the code base, associated tests and fill gaps in those tests, thereby grilling the code. Then ask it to work on what you intend which is related to that part of the code. I've had some very good success this way.

Important lesson for using LLM's well: If you want to produce a document (a PRD, an article, a long email), split it into two phases: 1. A loose grilling session where you align on what you want to create (/grill-me) 2. The creation of the document (/create-doc) If you go with /create-doc from the start, the LLM will be far too eager to create the doc and won't ask crucial questions. You'll end up with something very different from what you wanted. And that crap first draft will bloat context for the rest of the session. (this, by the way, is why plan mode sucks. Ideating and creating the plan should be two separate steps)

@mattpocockuk What theme is this? It's poppin crisp

This is how it would look:

Thinking about moving Sandcastle off Docker and making the sandbox totally pluggable. I.e. Sandcastle would become an orchestrator that works with any coding agent in any sandbox - local or remote. Would love feedback here: github.com/mattpocock/san…

@sarah_edo Very cool. Congrats! 🎉

📣 After nearly 5 fulfilling years leading Core infrastructure organizations for Google, I’m thrilled to share that I’ve taken on a new role! I’m going back to IC work as a Distinguished Engineer- my new role is in Chrome as Area Tech Lead, AI Web Ecosystem! 💥 The rise of AI is driving a fundamental sea change in web architecture. While this era promises immense innovation, it demands rigorous focus on security, privacy, and user observability. AI is a technology. It must be integrated with the same deliberate care as the major technological shifts that preceded it. In this new role, I’ll be working alongside world-class colleagues to redefine the web platform for an AI-first runtime. My mission is to architect a composable web that enables safe, standardized, and mutually beneficial interactions between creators, users, and AI agents. With this transition, I am moving from leading large organizations back to the IC track. I’ve swung this "managerial pendulum" several times in my career, and I’m energized to return to deep technical work at a moment that so clearly demands it. I'm also excited to be working with/for web legends like @laparisa @jimbankoski @paul_irish @RickByers @Paul_Kinlan @groby @mathias @bmeurer @fractorious and so many more! 💥

@Hamlinsolo18 @cenkuygur Receipt? I'd like to see that...

@cenkuygur You literally told people he was the better option than n Kamala... fucking fat retard

I’m the last person to rub it in to people who voted for Trump. I’ve been saying from day one that we should be open to our fellow citizens who were misled by him. But today you can see why I voted for three corporate Democrats I can’t stand over him. He has always been a madman.

@mattpocockuk What about a dont-use-react skill? 🤔

Every React app should probably have a 'dont-use-use-effect' skill That explains why you might not need a useEffect, and what state/effect manager you use instead

@cenkuygur I thought the last weapon you have would be a general strike.

American people seem to have no power at all. No matter how clear we are or how united we are, politicians and national media always support the donors over us. We have one last weapon left - voting. Not for one party or another, but against all the incumbents who’ve betrayed us.

@davidfowl Looks clean. What's the magic? It does a vite build for deploy by convention?

The next version of aspire will support TypeScript based apphosts and without .NET being installed on the machine. We prototyped ~4 different approaches over the last year until we found a path that we liked. Here's a react (with vite) + express app described using. #aspire #typescript

@Principles_1st @MarkHertling You can get some of it back by drops, but the rest of the world will never trust you like it did. The trust roof has been lowered permanently.

.@MarkHertling: "Trust is gained in drops and lost in buckets. We've lost a tanker truck full of trust over the last year, and the only thing we can do to rebuild it is replace it drop-by-drop. And I believe we will."

@davidfowl I do, but I think it will plateau faster than what you think. I guess we'll see. One thing to agree on: it's exciting times. 😊

@GauteLoken You don’t think we’ll get better at less steering? It’s a skill

Ultimately we all want to get here, and the code quality will get better, but right now multiple passes over the code are needed to keep the quality high.

@davidfowl Not sure I can see how much steering there is, so I can't guage how autonomous those contributions are. I use AI a fair bit myself, but for now I'm steering and checking and iterating for results to be good. At that point, AI may be writing much of the code, but I remain author.

@GauteLoken You know .NET is open source, you can see how much AI is being used to develop it. It's insane :)

@davidfowl That's not getting there, then. But glad to hear you're somewhat saner than what your comment indicated. I would hate for. NET to become trashy because of reckless AI use.

@GauteLoken We'll build systems to make sure the quality of the code is good, that might not be everyone's job in the future though

@evanyou 👀

Working on a new idea… I haven’t been cooking like this in a long time.

@mattpocockuk Ah.. And then I read your second tweet. I see, you're not holding your breath either. 👍

@mattpocockuk Just cause you confidently assert that they will soon be good enough to counteract software entropy on their own, doesn't make it true. It might be true, but I'll not hold my breath while we wait for that day.

A possible future: For the last 30 years, devs have made the sensible assumption that every time a human touches a codebase under time pressure, the codebase gets less maintainable. This is called software entropy. It's only possible to escape by continual monitoring and improvement. Or by employing ONLY 10x devs (impossible). The current generation of LLMs aren't currently good enough to escape software entropy. Without careful supervision, they make codebases worse. But soon, they will be good enough that they will make the codebase better each time they touch it. At that point, we won't need human review any more. Codebases will be better as a black box. We'll review inputs and outputs. Before Opus 4.5, models produced software entropy at an alarming rate. Opus 4.5 was the first time I felt like the entropy was manageable. Soon, there might not be any entropy at all.

@davidfowl You were so much cooler when you weren't in the hype train...

Upgrading code bases, issue triage and reproduction, building the first cut of features, doing deep code reviews, performance testing and analysis, all while you're asleep.

Coding agents are going to be used at scale to automate so much software engneering tedium. It's going to happen this year, just wait.

@grok @InAbundance99 @cenkuygur Would it not make sense that when an ally is proven to interfere in the elections of the US, their status should change from ally to enemy, and so what constitutes treason also changes?

Treason in the US is narrowly defined by the Constitution as levying war against the US or giving aid and comfort to its enemies. Israel is a US ally, not an enemy, so alleged coverups to protect Israeli interests wouldn't qualify. Recent Epstein files show ties to figures like Ehud Barak, but claims of intelligence ops are disputed—Netanyahu denies them, and experts warn of conspiracy theories. If proven, it might involve other laws like espionage, but not treason.

Release of the Epstein Files has become the lifting of the curtain in the Wizard of Oz. Now, we see everything has been a mirage all along. I think the coverup has been more revealing than the crime. To this day, no one in national media will tell the truth we're all seeing with our own eyes - it was Israel. At every turn, he's trying to help Israel. The intelligence agencies definitely know. They're the ones who were protecting him. There is absolutely no question of that after reading through the files and the @DropSiteNews articles. So, why are they all protecting them and lying to us? Why are our intelligence agencies helping them? Why is our government protecting them? How long has the media been in on it? Both parties knew it. And they hid it from us. How much power does Israel have inside our government that it can force any US president to do as they wish? It appears they were all told to cover up these terrible crimes - and they all did. Every president should be asked - you undoubtedly covered up these crimes, why did you do that? And who asked you to do that?

@day9tv Play something with @followgrubby! 🫡

YEAR OF RTS BEGINS! In 2026, I'll be streaming tons o' RTS games, new and old, multiplayer and campaign, indie and AAA and everything in between! Today? We start with the Command & Conquer Red Alert 2 SOVIET campaign! Twitch -> day9tv

@MarkHertling I fear we Europeans may have to repay the favour in a year or two - travel to the USA and help drive out a fascist regime. For now I'm cheering for you to sort this out on your own. Good luck! 🫡

I’ve had these same kinds of experiences during my time in Europe. The strongest was at the Netherlands American Cemetery, where local citizens adopt US soldiers graves to “teach their children” about American sacrifice for fellow citizens. It’s why many of us are so angry at what we see this administration doing.

@talb_bluecurve @jcouv That's going to make it confusing to javascript+C# devs. Good feature but pick a different keyword.

@jcouv Nice but we need an extension of the let keyword so we can define readonly variables, so we will have var and let. It will allow for most robust code.

We merged a C# 15 feature into .NET 11 preview 2: collection expression arguments. Allows specifying creation arguments for a collection expression: `List<string> names = [with(capacity: 10), ... elements ...];` Spec: github.com/dotnet/csharpl…