This weird trend reached an apex in a Feb 2026 OpenAI blog post [1], recently on the front page [2], which describes the process for building... something... written 100% by agents.
There is no description of what the thing is, no indication of what value it provides its users. The closest it gets is "the product has been used by hundreds of users internally, including daily internal power users".
But the fact that the thing has a million lines of code is repeated twice in the first few hundred words.
> "the product has been used by hundreds of users internally, including daily internal power users".
My guess is it’s an email filter.
> million lines of code
> written 100% by agents
Yeah, probably an email filter. Or maybe a JS menu for a departmental wiki that basically recreates jquery using MS JScript and transpiles it into JS 5.
The email filter team is trying to match the pace of innovation of the email generation team. At stakes is the ability for the employees to process the billions of mission-critical generated emails each of them receives each day.
The best strategy is to figure out what you are going to do after smoking weed before you smoke weed. So, draft your prompt and send it before lighting up.
The entire Linux kernel is about 40 million LoC, and only something like 16 million LoC after you remove drivers. I have a hard time imagining whatever OpenAI was talking about there having anywhere close to 6% as much utility as the Linux kernel, despite having 6% as many lines of code. And I have a hard time imagining it's anywhere close to maintainable, regardless of how powerful their LLMs might be.
To be fair, few things of any number of LOC have as much utility as the Linux kernel, and it's also a particularly dense example of code. There's plenty of other examples that have higher LOC / utility ratio without being vibe coded. For example, Google's monorepo famously has 2 billion LOC, which is a statistic I've heard long before LLM coding took over.
Clarification: Google claimed to have 2 billion lines of code in their repo ten years ago, and a commit rate of 50,000 changelists per day, both on exponential growth trends.
That's certainly a way to look at it. And that repo contains a "third party" directory which itself contains Linux, LLVM, and much of the rest of the open source world. But I would suggest that the largest of those thousands of applications probably has a transitive closure of hundreds of millions of lines of code.
I'm aware. I worked on that specific project (assuming we are talking about the same one) back in the day. :-)
There are certainly very large applications in that repo in the hundreds of millions of lines of code. But comparing the entire repo to single applications is not an apt comparison.
The Linux kernel is not in any way at top of big projects. A kernel, as the name suggests, deals with specific issues and tries to remain small.
The world’s biggest software is usually built over endless adapters of different data and a need to reconcile endless edge cases with laws, regulations and real world complexities.
When I was young, I remember a (joke) program from a book or magazine, called something like report writer or something. It was written in basic and you typed it in.
You would run it and it would say:
how many pages? _
You would type in a number, and it would generate that many pages of a complex-sounding report.
something like "the subsystem design interface is ..." blah blah
The reality distortion field is strong around anthropic. Anthropic posts tons of equally bullshit blog posts, written entirely by AI, saying absolutely nothing, to the front page and they consistently average many hundreds of upvotes.
I have to wonder if HN is astroturfed. Anthropic and OpenAI can post about a new model and it gets over to HN instantly and is full of glowing reviews of supposedly real people who supposedly have had prior access to it who supposedly think it's the best model yet.
Feels doubly so for Anthropic. There's just unbelievable level of glaze, and insane upvote velocity for "blog posts" that are essentially fluffed up feature documentation.
Somehow everything boris says has become the word of God. The dude is just an engineer, like you and me, who gets unlimited tokens for free.
Yeah, I sometimes catch these posts minutes after they were submitted and they're full of highly upvoted glowing reviews that maintain their momentum on the top of the page.
And this is hacker news. A place where famously the most upvoted comment is the one critical of the post.
Upvoter here. Nothing special -- I like Claude a lot. I found it more ergonomic to work with, and I think its models are state of the art. I believe I'm more productive especially in legacy crufty codebases.
I may not be representative of the universe or have a controlled, randomized study to back it up, but that's not what upvotes are for are they?
That's a fair point. To play devils advocate to myself, it's possible that the huge upvote share is due to a much bigger marketshare among our community.
Yeah, that was my guess too. Still a little disappointing to see fellow HNers reflexively fanboying a company that's the overwhelming dominate player in LLM coding.
I try to reserve my reflexive fanboy company/project votes for underdogs who need and deserve the help.
The tech field has always been full of naive technology boosters. HN might be heavily astroturfed for all I know but I am sure there are many real people in a state of constant excitement over the progress of AI.
Heavily. There’s an amount of true believers who actually think RSI is going to happen any day now, but you can spot patterns amongst brand new booster accounts…
I real life I meet people who like AI and people who hate it, but nobody who’s on a personal mission to defend Anthropic from anyone that dares to question their hyperbolic marketing.
HN is probably astroturfed, but it is also one of those places where a lot of actual subject experts frequent. I don't find it unusual that a new model gets instantly posted here. It is also not unexpected that the most recently released model is the best model.
Yeah it was very disappointing that so few details were provided. One of the reasons I think it's going to be an open source project or effort that is going to show, sooner or later, how effective these things actually are.
In their podcast interview, they mention that it's an Electron app that users download, and so they periodically create a new build. See section "Autonomous Merging Flow" here: https://www.latent.space/p/harness-eng
Just wanted to check in on how we're tracking, I've booked a quick sync meeting for 4:45pm on Friday followed by a release cadence checkpoint meeting for 9am on Monday.
We've known for decades that output metrics like LOC/day are very bad measures of real productivity in software. But they seem to be back in vogue in the age of AI, because AI is so good at maxing these useless metrics, and we need to show how impressive our AI is and how impressive our usage of AI is.
This post and TFA have a common issue: no one seems to have a clear, compellingly evidenced account of basic questions about the collection and its history under consignment:
1. What exactly was in the collection?
2. What happened to the collection after it was consigned: which sets were sold, which were stolen or lost, which were moved to off-site storage, etc.?
3. How much money did the original franchise owner owe the consigner for the sets sold?
The peripheral claims about e.g. police malfeasance are disturbing, but without this basic evidence about the substance of the matter, I don't know if it's a great idea for an online mob to take sides.
One thing really stands out, which is that Bricks and Minifigs is pointing the finger at the franchisee for an "unauthorized" consignment deal. While the blog post has a screenshot of the contract specifically allowing consignments. Which undercuts a large part of the corporate argument.
There was a livestream last night where the Bricks and Minifigs COO said that the franchise agreement has a clause saying that the operations manual overrules the franchise agreement and they can update the operations manual at any time. BAM's court filing instead says that the franchise agreement allows consignment services, but "limited expressly to those approved by BAM".
There's good evidence that it was expressly approved in the form of social media posts advertising the consignment on their social media pages. Difficult for them to argue either ignorance or that the arrangement wasn't authorised, both of which aren't really relevant to the demonstrated facts of the dispute.
You're right that we don't have the facts. That said, the way that Bricks and Minifigs handled this has been horrible. The CEO should have said on camera that he's willing to work with the lego owner to return the property and to work with police to charge the former franchise owner with theft.
You're saying that people shouldn't take a stance "without this basic evidence about the substance of the matter". The basic evidence is there. Take the time to watch Ben's videos, it's an hour of your time well spent.
What I keep wondering is, what would have happened without the AI? Would they have just ignored your request?
In my neck of the woods it's fairly common that when a person doesn't know how to help you they just don't reply, instead of saying "I don't know how to help, sorry". AI-generated responses seem like the evolution of this attitude that one must either ignore or respond in a (superficially) helpful manner.
Yes, this and every internet forum will still be doing this two years hence. Your life will be better if you take to heart this famous passage from Nietzsche:
I do not want to wage war against what is ugly. I do not want to accuse; I do not even want to accuse those who accuse. Looking away shall be my only negation.
I’ve been thinking of building myself my own frontend to HN that makes it impossible to view comments, for this reason. Yet sometimes there are still really interesting discussions and it’s hard to let go of what for me feels like the last social media I want to be part of.
I think they're saying it has qualitatively different capabilities that make certain kinds of security work more worth pursuing with the model, not that the model of human-AI interaction has changed.
You're right that they're using a harness like everyone else. The general idea of giving the model a harness is not going to change. I mean even humans need harnesses to accomplish some things.
Well they don't say they didn't read the thing. But 10-20 minutes often isn't enough time to truly digest the arguments. And even it were, I'm not sure it would matter. What I've seen is 10-20 minutes of awkward silent time followed by a level of discussion that isn't notably better than meetings before we adopted Amazon ways. Probably we are doing it wrong and/or have the wrong people in the room.
10-20 minutes seems short. Most doc reads I’ve been in at Amazon have a 30-40 minute reading phase. 10-20 minute reads would only be for very short or simple documents.
You joke but that's pretty much what cross functional teams are.
The only other observation is that as you grow teams, communication channels multiply exponentially and at over 6-8 people communication starts breaking down.
So instead you make small "companies" and set a few ground rules which software they build needs to follow, and you are back at a working org producing complex software.
Putting everyone responsible for some function of a product on one team, instead of having separate departments for separate functions, can do wonders for actually shipping and iterating on software.
I don't remember who said it first but "software teams always ship their org chart" has always stuck with me. How a team is organized has such an outsized hand it what they build.
There is no description of what the thing is, no indication of what value it provides its users. The closest it gets is "the product has been used by hundreds of users internally, including daily internal power users".
But the fact that the thing has a million lines of code is repeated twice in the first few hundred words.
[1] https://openai.com/index/harness-engineering/
[2] https://news.ycombinator.com/item?id=48416264
reply