Sean Goedecke

I think you're being a bit harsh here. Yes, the blog is mostly about AI, and yes, he publishes very frequently. But his articles don't read like AI and he claims not to have used it in his writing (https://www.seangoedecke.com/avoid-ai-writing/). And regardless of how you feel about the content, the community has clearly decided it's worthwhile as a discussion point.

Absolute classic! https://www.seangoedecke.com/static/3c8f2a6459ed23310c4eb51d637fe007/4…

> I still believe those who rave about it are not writing anything I would consider "engineering". Correct. In fact, this is the entire reason for the disconnect, where it seems like half the people here think LLMs are the best thing ever and the other half are confused about where the value is in these slop generators. The key difference is (despite everyone calling themselves an SWE nowadays) there's a difference between a "programmer" and an "engineer". Looking at OP, exactly zero of his screenshotted apps are what I would consider "engineering". Literally everything in there has been done over and over to the death. Engineering is.. novel, for lack of a better word. See also: https://www.seangoedecke.com/pure-and-impure-engineering/

After his first #1 HN post on Nov 11, 2024, Sean Goedecke went on a ridiculous posting run in 2025. Over 130 published and 2nd only to Simon Willison. 1. http://simonwillison.net — 21,015 points 2. http://seangoedecke.com — 11,243 points 3. http://jeffgeerling.com — 11,106 points 4. http://neal.fun — 8,998 points (374.92 point per post!)

> "Even GPT-5, Claude 4.5, and Gemini 3.0 agreed: This is the kernel of AGI." Of course they did, they are sycophants They will tell you any idea is a great idea. It has no bearing on the actual "goodness" of an idea https://www.seangoedecke.com/ai-sycophancy/ (currently on the front of HN) > the LLM executed it as a game engine Did it? How did you verify? How do you know it's not just telling you it did something without actually doing it?

There was a blog post the other day arguing that great developers aren't 10x but infinity-x. It sounds ridiculous, but their point was that great developers don't simply program faster but solve problems other developers are entirely unable to solve. And that makes sense to me. It's the difference between an AI boot camp graduate and someone with an ML PhD. They might apply a sklearn function at the same speed, but as soon as complex problems appear, the former hits an unsurpassable wall. edit: https://www.seangoedecke.com/weak-engineers/

Another post from the same author puts this in an interesting context: https://www.seangoedecke.com/glue-work-considered-harmful/ (follow up: https://www.seangoedecke.com/cynicism/) Keeping the code base tidy is glue work, so you should only do enough of it to ship features. So maybe these are not "mistakes" but rather tactical choices made by politically smart engineers focused on shipping features.

Re productivity: the METR study is seriously flawed overall, and: 1. if you disaggregate the highly aggregated data, it shows that the slowdown was highly dependent on task type, and tasks that required using documentation or novel tasks were possibly sped up, whereas ones the developers were very experienced with were slowed down, which actually matched the developers' own reports 2. developers were asked to estimate time beforehand per-task, but estimate whether they were sped up or slowed down only once, afterwards, so you're not really measuring the same thing 3. There were no rules about which AI to use, how to use it, or how much to use it, so it's hard to draw a clear conclusion 4. Most participants didn't have much experience with the AI tools they used (just prompting chatbots), and the one that did had a big productivity boost 5. It isn't an RCT. See [1] for all. The Anthropic study was using a task far too short to really measure productivity (30 mins), and furthermore the AI users were using chatbots, and spent the vast majority of their time manually retyping AI outputs, and if you ignore that time, AI users were 25% faster[2], so the study was not a good study to judge productivity, and the way people quote it is deeply misleading. Re learning: the Anthropic study shows that how you use AI massively changes whether you learn and how well you learn; some of the best scoring subjects in that study were ones who had the AI do the work for them, but then…

This is a great explainer from an LLM perspective, and it would be interesting to see a computational scheduling explanation in depth. I presume that hyperscale LLM companies extensively examine the computation trace to identify bottlenecks and idle bubbles, and develop load balancers, pipeline architectures and schedulers in order to optimise their workload. The batching requirement for efficiency makes high security applications quite difficult, because the normal technique of isolating unrelated queries would become very expensive. The nVidia vGPU GPU virtualisation time shares GPU memory, and every switch requires unload/reload context switches, doubtful they have deduplication. Multi-Instance GPU (MIG) splits GPU memory between users, but it is a fixed partitioning scheme (you have to reboot the GPU to change), and nobody wants to split their 96GB GPU into 4x24GB GPUs. Makes me wonder what the tradeoff is for putting second level memory on the GPU board (i.e. normal DRAM), so that different matrix data can be loaded in faster than over PCIe, i.e. the HBM becomes a cache. (I'm also really liking the honesty in the authors book on Software Engineering, not in the dry IEEE sense, but as a survival guide in a large enterprise. https://www.seangoedecke.com/book/ )

From my understanding, AI outputs have patterns due to its text prediction algorithm and AI detectors just recognize those patterns as watermarks. If there used to be 1,000 different ways to write a paragraph, watermarks may reduce it to 300 and that’s still a lot. Humans shouldn’t be able to perceive these changes except the output is really short. https://www.seangoedecke.com/text-ai-watermarks/

Would you side with the adjacent statement "expertise is not the bottleneck"? Didn't we recently discuss "LLMs reward expertise"?[1] I suspect Discovery Loop will have to hire experts in each area they are targeting, to supervise and prompt their system effectively, much like the Terence Tao conversation with ChatGPT the OP cited[2]. [1] https://news.ycombinator.com/item?id=49161518 [2] https://www.seangoedecke.com/llms-reward-expertise/

there's also a DeepSeek whitepaper on this technique https://www.seangoedecke.com/text-tokens-as-image-tokens

Yeah you definitely have to be skeptical regarding sentiment for open/local model capabilities, since there's bias from what people want to be true. I generally agree with this in spirit https://www.seangoedecke.com/are-new-models-good/ , but I think you can read Anthropic's results showing Sonnet 5 as almost strictly worse than Opus 4.8 as very credible/meaningful, and then draw comparisons from that

Middle managers were about making work legible https://www.seangoedecke.com/seeing-like-a-software-company/ Now that wasn’t the most efficient and it has a whole bunch of perverse incentives too, but it’s what happened as companies got big enough to hire people to do the work that could be delegated because those delegatees got status from having people to themselves delegate to.

no i am talking about a page like this - https://www.seangoedecke.com/page/2 or - https://mmapped.blog/posts this or - https://kevinpatel.xyz/posts/ this

Why are we upvoting this? Virtually, the entire blog is about AI with a ridiculous publishing rate (https://www.seangoedecke.com/page/5), funny how I can look at this site HTML and know right away it was done with AI. Can we stop upvoting vibe published articles? The arguments are flawed and don't even make sense to anyone who does software

it is not. the real world says one thing and does another. here is how real world tech companies actually function: https://www.seangoedecke.com/how-to-ship/

https://www.seangoedecke.com/generate-skills-afterwards/ - the response

Many such cases. https://www.seangoedecke.com/gas-and-ralph/ (https://news.ycombinator.com/item?id=46654878) https://news.ycombinator.com/item?id=46777411 https://news.ycombinator.com/item?id=46780065

>Shipping is a social construct within a company. Thanks for flagging this, this was an epiphany for me today, so for anyone else struck by it I'm linking directly to the article it's from (same author, and linked from the article in the context the parent mentions, just not linked directly in their post above): "How I ship projects at big tech companies" https://www.seangoedecke.com/how-to-ship/ Also the HN comments on it from when it was originally posted: https://news.ycombinator.com/item?id=42111031

Re: the plot, there's also this one which fits in among those two: https://www.seangoedecke.com/shareholder-value/

There was an interesting discussion on here a couple of months back about images vs text, driven by this article: https://www.seangoedecke.com/text-tokens-as-image-tokens/ Discussion is here: https://news.ycombinator.com/item?id=45652952

I think Blow and Muratori are pure engineers (as defined here: https://www.seangoedecke.com/pure-and-impure-engineering/). Pure engineers deliver perfect and fast software somewhere along the Black Hole Era. Not quite heat death of the universe, but almost there. Impure engineers deliver "working" code in a deadline, for an arbitrary definition of working. Basically, The Worse is Better™.

Are you based in the US? I've heard that compute capacity peaks during US working hours, and access may be degraded during that time, for example, through dynamic quantization [1]. [1] https://www.seangoedecke.com/ai-is-good-news-for-australian-engineers/

For those here who are dismissive of the value of the humanities, consider that no problem and no solution is purely technical; there are always "humanistic" aspects. One can - and many do! - ignore these, or even be totally unaware of them, but they're there to be understood all the same. If you're curious what I mean by this, Sean Goedecke's post "How I Ship Projects At Big Tech Companies" [1] is a superb example, particularly his definition of "what does it mean to ship?" No idea whether he's somebody who would say "the humanities are important" but I don't think you can understand his thesis as a technical one. [1] https://www.seangoedecke.com/how-to-ship/

See also: Do the simplest thing that could possibly work https://www.seangoedecke.com/the-simplest-thing-that-could-possibly-wo…

today I also read this, and I find it related: https://www.seangoedecke.com/autodeck/

This is very clearly written by ChatGPT. See: https://www.seangoedecke.com/chatgpt-house-style/

And the reason they're all in on streaming to begin with is because they're sending massive amounts of data back and forth all the time. Like Sean Goedecke said in his API design writeup [0], a technically poor product can make it nearly impossible to build an elegant API. I believe we're seeing the same thing with Next.js, all of these wonky interfaces derive from the underlying architectural issues. [0] https://www.seangoedecke.com/good-api-design/

>https://www.seangoedecke.com/programmer-archetypes/ I found this one super interesting. Personality types that make sense but I haven't seen them described this way before.