Hacker News

points by gradus_ad 3 days ago | 0 comments

The killer app is AI coding. Claude Code had first mover advantage, was a genuine revolution, and has diffused through the industry with incredible speed. I would imagine most of the huge revenue increase over the last year is due almost entirely to AI coding spend.

There are three issues with this. First, the rapid diffusion of the technology means it is very unlikely to continue at the rate of adoption seen over the past year, so it's very hard to extrapolate based on what we've seen so far. Second, many companies are reigning in AI budgets after an initial honeymoon phase. Third, all of the other labs are trying to take a bite of the coding pie, and the price war is already heating up.

Add these simple dynamics up and it makes perfect sense why Anthropic would want to IPO now. Things look about as good as they ever will.

[–] random__duck 3 days ago

Just like Alta Vista and Netscape put there first mover advantage to use?

[–] jfoster 2 days ago

That's essentially the point. There is a non-zero risk that any given AI company ends up like those examples. That's why they are racing to IPO, raise more, get ahead of each other, and reach escape velocity of that risk.

[–] toephu2 3 days ago

Claude Code is a commodity at this point though.

Codex is just as good if not better. And theres 3-5 other options out there just as good too (harness + model).

[–] deskglass 3 days ago

Opus 5.5 is the best deal in doing rn. It will change soon. But its not the case that codex, etc are equivalent

[–] ksh09 3 days ago

I don't use claude-code and like to use the harness of my choice.

[–] dbmnt 3 days ago

I disagree. I've tried Codex, and two of the provider-agnostic harnesses (Pi and OMP), and I strongly prefer Claude Code.

[–] alexfortin 3 days ago

I disagree. I've tried Claude Code, Codex, OpenCode and Pi, and I _strongly_ prefer Pi.

I suppose it truly boils down to personal taste? My take[0] on AI including harnesses.

[0]: https://a.l3x.in/ai

[–] ffsm8 3 days ago

piagent vs Claude code are completely different target audiences...

If you drew a venn diagram of their prospective users, the overlap ain't gonna be big. I get that they're both harnesses, but claude is clearly trying to optimize for the agentic development loop, while piagent is more about hand holding - unless you add the agentic features yourself (which is pretty easy, but again completely different target audience to cc).

[–] alexfortin 2 days ago

Actually the main reason to use Pi is because it's lean and doesn't pollute the context with all sorts of noise like most of the $corporate agents do.

A nice benefit of going minimal with coding agents is that you might find out you don't need to do any hand holding because nowadays models are perfectly capable of reaching out for tools etc and make you happy without any extras required. In my experiece so far less truly is more when talking about agents.

[–] HighGoldstein 3 days ago

"I prefer Claude Code, so it must be the best"

Truly a developer mindset.

[–] jimkleiber 3 days ago

Well, plenty of businesses have been built on not being the best but being the most preferred regardless.

[–] rienbdj 3 days ago

Yeah why?

I bounce between codex and cc all the time. There is no moat there.

[–] dizhn 3 days ago

I tried the Claude $20 plan for two weeks and burned the rest of it due to how much token anxiety it gave me. And I'm not ready to keep throwing money at it for it to become usable. Gemini models are actually fine for me as an amateur+ I found especially since they have regional pricing and a very generous side items package.

I use the models but not the "harness" though. Not directly anyway. I am still handicapped by the CLI when running in Paseo. Both antigravity and agy are toys compared to what has been available for at least two years.

[–] Degorath 3 days ago

That's quite surprising, of the set {Claude Code, Codex, Pi, OMP} I must say I find CC to be the worst.

They all have some annoying downsides and I use CC the most, but it's gotten to the point where I'm for sure actively shopping around for better harnesses.

[–] WinstonSmith84 3 days ago

Huh.. well maybe... but I can only assume you are not a developer

[–] NewJazz 3 days ago

Why though?

[–] noname120 2 days ago

I’ve been a loyal Pro 20x subscriber for 10 months but I plan to switch to Opus 5.5 very soon because both the 6.1 Sol and Astra models severely underperform, there is just no good reason anymore to stick to them.

[–] parthdesai 2 days ago

Majority of the people at my company switched to Codex. All it goes to show is that performance difference at this point is minute and is a matter of taste

[–] noname120 1 day ago

That was true before Opus 5.5, not anymore

[–] threetonesun 3 days ago

View them as a successful SAAS company deeply embedded in tech company processes like Atlassian. I don't see Anthropic or SpaceX actually having a market on all the software in the universe but I don't know that the market won't buy them like they do.

[–] potatototoo99 3 days ago

My brother in christ, there is no first mover advantage. I use claude code daily and if the open ai copies it 1$ cheaper per seat my company is switching immediately.

I've never heard anyone in real life discussing what model is better in some benchmark. Money talks.

[–] tclancy 3 days ago

That last sentence is a great point. We are probably in a bubble here just by virtue of regularly visiting HN. Even at work in an AI department at what I think is a typical company, there is little to no opinion on models beyond correctness and cost.

[–] tim333 2 days ago

>Things look about as good as they ever will.

That kind of assumes things are static and Anthropic won't get a first mover advantage in some new area. If you look at the recent progress in maths it seems unlikely AI has plateaued.

[–] 3RTB297 2 days ago

I think their comment was more about user and consumer movement, plus both desire for, and need to utilize, frontier models to get the task done.

With the token budget sticker shock and end of the honeymoon phase, just a guess, but I expect a significant market share might start to look at open weight "not quite frontier" models and outsource compute alone and have greater control over the model rather than outsource everything. Which will also be a trend that peaks and folds as it's discovered it takes a lot of human work to get what you need. Maybe then settling into a pretty standard tier of medium-quality contractors that specialize in LLMAAS for cheap, running open weight models.

Which isn't to say that AI has plateaued, but rather that budget-friendly, functional options might undermine the importance of frontier models sooner rather than later.

[–] smashed 2 days ago

I may be dumb but I don't see how progress in math could bring billions in revenues like progress in coding did.

[–] tim333 1 day ago

I was thinking more that general progress in AI may lead to competent AI lawyers doctors managers etc.

[–] codingisinteres 2 days ago

I find the perspective of outsiders on "coding" really interesting. Recently I heard Paul Kedrosky make up some cool-sounding analysis of how "coding" is perfect for "next token generators" but "finance" is really something else. His argument boiled down to "coding is inflationary" and "finance is deflationary": the prompt "make fitness app" expands into a million LoC and the prompt "analysis this financial statement and tell what's interesting" does the opposite (or something). It's insane how people with influence can have opinions that are disturbingly irrelevant at best and flat out wrong at worst.

It's hilarious to watch non-coders have opinions on us while once I form an opinion on someone's work it's somehow invalid. No, "coding" is not just "write me code". I ask it all the time to "tell me how this system is designed, how it interacts with its environment and tell me any points of particular interest". It's the same bloody process just on different concepts. That's actually what it's really good at. Your "finance" job, while I'm sure is very complex (sigh), does not contain divine elements that are somehow beyond the grasp of correlation itself. Whence do these magic elements spawn from? (The heavenly abodes?)

To write "code" effectively you have to understand your environment and clearly grasp the unstated but very present background radiation of not only your particular business but dev and even tech culture in general. The models are becoming scary good at this. I feel it's very irresponsible to keep harping on "code" as something that's somehow special. It's absolute not. It's just very, very high leverage: automate automation and you got it all. No other job has that property except perhaps AI research itself but which itself again needs code to materialize.

These models are being RL'd to death on code. I wonder what happens once some off-shoots will finally be RL'd to smithereens on law, medicine and finance. I however wouldn't start with that. Code is much more valuable because it is generative and high-leverage. There is no end to what you can achieve once you crack that. Law and medicine are uninteresting specks of dust compared to that. They are closed, solvable problems in comparison. What if "crack" "law"? This buys you absolutely nothing. You can't code up a robotic development platform with law or medicine. You can't setup infrastructure with law. It's useless for our purposes.

[–] epsteingpt 3 days ago

underrated comment

[–] metalspot 3 days ago

> Things look about as good as they ever will

Anthropic looks like Wylie Coyote spinning their legs in open air at the moment. They are fucked from so many angles it is hard to comprehend how anyone would buy their bags.

Simple question: What is their IP? What are you actually buying? They didn't invent anything. They have no IP. They copied other people's IP and did a mediocre job of it.

There are a half dozen Chinese companies who can build the same thing, better, for a tenth of the price, and will give it away just to fuck with the US. If you are buying Anthropic you are buying a ticket to a hurt box where you are going to get relentlessly pummeled from all sides until your capital turns to ash.

On top of all this, the company is run by an incompetent psychopath and a bunch of dumb cultists, who are blacklisted by the US Military, and have at least another 2 years of the current admin doing everything possible to cripple their company.

What is the story for how you succeed in the most important strategic technology of all time when you are blacklisted by your government's military? There is no story because you simply don't exist in that scenario.