Andrew Hall

@davidclowery I too will not be performing, for which I apologize.

I will also not be performing at the Great American State Fair. I was never asked. I doubt anyone that books events at this level has ever heard of me. So it shouldn't be a surprise. But I apologize anyway.

@IceSolst Compilers are deterministic functions written and owned by humans with copious prior testing. The output has been well vetted before I ever use it. If there is a bug, a human who knows the code will fix it. This is not the case by any measure with AI output from a prompt.

Interesting article on treating agent output like compiler output (and why) skiplabs.io/blog/codegen_a…

@larsencc None of this is particularly relevant - soon there will be no technical advantages. It will come down to service excellence - do you serve your customers so much better and efficiently than everyone else?

Hot take: the moat in software isn't your code or how fast you ship anymore. Everyone can code and code fast now. The moat is how well you wire up the loops and processes. Hooking up your entire stack (e.g. Datadog, AWS, Terraform, repos, Slack...) to agents that can monitor, find bugs, fix them, and deploy 24/7. That's hard (at least for now). Making those pipelines reliable and safe at scale is the actual engineering challenge now. The companies that nail these end-to-end loops first will be genuinely impossible to compete with. At @browser_use we are making bets.

@handre "population of 320,00" - that is the same size as Tampa, FL. Equivalent size failure in the US would be a local regional bank. Those go out of business without extraordinary intervention all the time. For example Washington Mutual Bank went under without a bailout

In 2008, Iceland did something unthinkable: they let their banks die. While America scrambled to save "too big to fail" institutions, this tiny Nordic island told their three largest banks to pound volcanic sand. The immediate aftermath was brutal. The krona collapsed 50% overnight. ATMs ran dry. And the entire country—population 320,000—found itself effectively bankrupt on the international stage. But here's the twist no economist saw coming: Iceland recovered faster than any other crisis-hit nation. By 2012, their economy was growing again while Europe wallowed in endless bailout cycles. They jailed 26 bankers, restructured household debt, and rebuilt from actual ground up. Meanwhile, we're still dealing with zombie banks propped up by endless Fed liquidity injections. Sometimes the "radical" solution is just letting capitalism work.

@mntruell What I don't think is being discussed is that we're in the era of constant greenfield. No one is interested in improving/refactoring existing systems. Everyone wants to start fresh. But what happens when these new systems become bloated. Start over again?

Either Elon Musk has programmed a new ad system where everything he likes is hyped in auto-generated posts under other people's accounts, or @Martina has been hacked. We really can't rule out the first possibility.

@Grady_Booch Fortunately I have ten years of experience with the technology already

@joewalnes Em dash is AI kryptonite

I've cracked it! A foolproof unbreakable human vs AI captcha test. --- Are you human? ◯ Yes - I'm definitely human ◯ Yes — I'm definitely human

Vibe Coding is eating our life. Consequences: carlmastrangelo.com/blog/the-end-o…

@Grady_Booch Let's call it "Computer Aided Software Engineering"

@bcherny @karpathy What is your policy on code comprehension? Do you allow code to be merge that human hasn't looked at or is confused by it but is submitting anyway because "it works"?

As always, a very thoughtful and well reasoned take. I read till the end. I think the Claude Code team itself might be an indicator of where things are headed. We have directional answers for some (not all) of the prompts: 1. We hire mostly generalists. We have a mix of senior engineers and less senior since not all of the things people learned in the past translate to coding with LLMs. As you said, the model can fill in the details. 10x engineers definitely exist, and they often span across multiple areas — product and design, product and business, product and infra (@jarredsumner is a great example of the latter. Yes, he’s blushing). 2. Pretty much 100% of our code is written by Claude Code + Opus 4.5. For me personally it has been 100% for two+ months now, I don’t even make small edits by hand. I shipped 22 PRs yesterday and 27 the day before, each one 100% written by Claude. Some were written from a CLI, some from the iOS app; others on the team code largely with the Claude Code app Slack or with the Desktop app. I think most of the industry will see similar stats in the coming months — it will take more time for some vs others. We will then start seeing similar stats for non-coding computer work also. 3. The code quality problems you listed are real: the model over-complicates things, it leaves dead code around, it doesn’t like to refactor when it should. These will continue improve as the model improves, and our code quality bar will go up even more as a result. My bet is that there will be no slopcopolypse because the model will become better at writing less sloppy code and at fixing existing code issues; I think 4.5 is already quite good at these and it will continue to get better. In the meantime, what helps is also having the model code review its code using a fresh context window; at Anthropic we use claude -p for this on every PR and it catches and fixes many issues. Overall your ideas very much resonate. Thanks again for sharing. ✌️

A few random notes from claude coding quite a bit last few weeks. Coding workflow. Given the latest lift in LLM coding capability, like many others I rapidly went from about 80% manual+autocomplete coding and 20% agents in November to 80% agent coding and 20% edits+touchups in December. i.e. I really am mostly programming in English now, a bit sheepishly telling the LLM what code to write... in words. It hurts the ego a bit but the power to operate over software in large "code actions" is just too net useful, especially once you adapt to it, configure it, learn to use it, and wrap your head around what it can and cannot do. This is easily the biggest change to my basic coding workflow in ~2 decades of programming and it happened over the course of a few weeks. I'd expect something similar to be happening to well into double digit percent of engineers out there, while the awareness of it in the general population feels well into low single digit percent. IDEs/agent swarms/fallability. Both the "no need for IDE anymore" hype and the "agent swarm" hype is imo too much for right now. The models definitely still make mistakes and if you have any code you actually care about I would watch them like a hawk, in a nice large IDE on the side. The mistakes have changed a lot - they are not simple syntax errors anymore, they are subtle conceptual errors that a slightly sloppy, hasty junior dev might do. The most common category is that the models make wrong assumptions on your behalf and just run along with them without checking. They also don't manage their confusion, they don't seek clarifications, they don't surface inconsistencies, they don't present tradeoffs, they don't push back when they should, and they are still a little too sycophantic. Things get better in plan mode, but there is some need for a lightweight inline plan mode. They also really like to overcomplicate code and APIs, they bloat abstractions, they don't clean up dead code after themselves, etc. They will implement an inefficient, bloated, brittle construction over 1000 lines of code and it's up to you to be like "umm couldn't you just do this instead?" and they will be like "of course!" and immediately cut it down to 100 lines. They still sometimes change/remove comments and code they don't like or don't sufficiently understand as side effects, even if it is orthogonal to the task at hand. All of this happens despite a few simple attempts to fix it via instructions in CLAUDE . md. Despite all these issues, it is still a net huge improvement and it's very difficult to imagine going back to manual coding. TLDR everyone has their developing flow, my current is a small few CC sessions on the left in ghostty windows/tabs and an IDE on the right for viewing the code + manual edits. Tenacity. It's so interesting to watch an agent relentlessly work at something. They never get tired, they never get demoralized, they just keep going and trying things where a person would have given up long ago to fight another day. It's a "feel the AGI" moment to watch it struggle with something for a long time just to come out victorious 30 minutes later. You realize that stamina is a core bottleneck to work and that with LLMs in hand it has been dramatically increased. Speedups. It's not clear how to measure the "speedup" of LLM assistance. Certainly I feel net way faster at what I was going to do, but the main effect is that I do a lot more than I was going to do because 1) I can code up all kinds of things that just wouldn't have been worth coding before and 2) I can approach code that I couldn't work on before because of knowledge/skill issue. So certainly it's speedup, but it's possibly a lot more an expansion. Leverage. LLMs are exceptionally good at looping until they meet specific goals and this is where most of the "feel the AGI" magic is to be found. Don't tell it what to do, give it success criteria and watch it go. Get it to write tests first and then pass them. Put it in the loop with a browser MCP. Write the naive algorithm that is very likely correct first, then ask it to optimize it while preserving correctness. Change your approach from imperative to declarative to get the agents looping longer and gain leverage. Fun. I didn't anticipate that with agents programming feels *more* fun because a lot of the fill in the blanks drudgery is removed and what remains is the creative part. I also feel less blocked/stuck (which is not fun) and I experience a lot more courage because there's almost always a way to work hand in hand with it to make some positive progress. I have seen the opposite sentiment from other people too; LLM coding will split up engineers based on those who primarily liked coding and those who primarily liked building. Atrophy. I've already noticed that I am slowly starting to atrophy my ability to write code manually. Generation (writing code) and discrimination (reading code) are different capabilities in the brain. Largely due to all the little mostly syntactic details involved in programming, you can review code just fine even if you struggle to write it. Slopacolypse. I am bracing for 2026 as the year of the slopacolypse across all of github, substack, arxiv, X/instagram, and generally all digital media. We're also going to see a lot more AI hype productivity theater (is that even possible?), on the side of actual, real improvements. Questions. A few of the questions on my mind: - What happens to the "10X engineer" - the ratio of productivity between the mean and the max engineer? It's quite possible that this grows *a lot*. - Armed with LLMs, do generalists increasingly outperform specialists? LLMs are a lot better at fill in the blanks (the micro) than grand strategy (the macro). - What does LLM coding feel like in the future? Is it like playing StarCraft? Playing Factorio? Playing music? - How much of society is bottlenecked by digital knowledge work? TLDR Where does this leave us? LLM agent capabilities (Claude & Codex especially) have crossed some kind of threshold of coherence around December 2025 and caused a phase shift in software engineering and closely related. The intelligence part suddenly feels quite a bit ahead of all the rest of it - integrations (tools, knowledge), the necessity for new organizational workflows, processes, diffusion more generally. 2026 is going to be a high energy year as the industry metabolizes the new capability.

@josecanciani @unclebob I would argue it was never about switch vs polymorphism per-se. but it was about effective modeling and design that led to one or the other. And the principle still holds in the era of AI. Maybe even more so. Good modeling and architecture IS the context to guide the AI.

I share the same skepticism with @unclebob. Readability has always been key since we spend more time reading than writing code. What I've accepted is LLMs stepping in as a primary "reader" and devs more as reviewers. So maybe we need to rethink what "readable" means. That doesn't mean we need to dogmatically follow every pattern just because it's SOLID or clean (and that was never the goal anyway). I think a lot of haters of design patterns jump at any little chance they get to criticize them Instead, we need to prioritize -even more- straightforward readability and understanding. Especially for AI. LLMs thrive on simplicity and conciseness; and they struggle with complexity and length, leading to more hallucinations. I remember the classic debate of polymorphism vs a simple switch statement, recently revisited by Bob himself and @cmuratori (github.com/unclebob/cmura…), both coming from slightly different angle. Casey argued for switches on performance grounds and simplicity. Now in an AI era, I'd say the simplicity of switches has a new strong argument in reading performance. Polymorphism can feel "magical", resolving things dynamically and requiring a big mental (or context window) model of the entire framework or application to fully grasp. And that's hard, for new developers and for LLMs. This is exactly the spirit I tried to convey in my last article on "Why I hate magic, a case (rant?) against Annotations" (x.com/josecanciani/s…), where I share some experiences on how over-relying on Annotations and magic code can complicate our understanding and debugging of our code. Being explicit and concise gives AI the clear context it needs to be effective, without the overhead of unraveling hidden behaviors. So I would answer: the secret hyper-productivity? Designing code that's inherently easier for AI to parse first, avoiding big context windows, while at the same time support the human in the loop. I'm not sold at all that we can ship production code without looking at it. Can it work? Sure. But there's so many issues AI can't manage yet, we are not even close for any critical piece of code.

@unclebobmartin Big if

The power to clean things up is just astounding! Using Claude (et. al.) should result in much cleaner code than any of us could have produced before -- if we are diligent!

@Grady_Booch Let's call it "Probabilistic Rose"

@unclebobmartin Well, we're handing over our industry to a bunch of languages models - what could possibly go wrong?

@developfwd Which I hope is not a sequence of living through past errors.

Horrible concurrency bug. Grok was sloppy and didn't protect concurrent variables properly. Took me hours to find. I haven't had one of those in years. Ugh.

@vlad_mihalcea "Chance favors the prepared mind"

Learning increases the probability of luck finding you

@3YearLetterman @DougJBalloon A lot will come down to how he financed his water bed.

@DougJBalloon This will be appointment viewing. I will come armed with PowerPoint slides

CBS is planning a series of ideologically diverse, non-partisan town halls treating questions like "How has feminism failed women?", "Why are liberals so smug?", and "What makes poor people so lazy?"

@BulwarkOnline Impressive reporting.

“All our work is being destroyed.” The FBI spent a generation relearning how to catch spies. Then came Kash Patel. bit.ly/4aggb77

@RamsNFL Fantastic win. If you are in LA and you're not a Rams fan, you are missing out on incredible football

Week 15 Victory Vibes!