I find it really strange that there is so much negative commentary on the _code_, but so little commentary on the core architecture.

My takeaway from looking at the tool list is that they got the fundamental architecture right - try to create a very simple and general set of tools on the client-side (e.g. read file, output rich text, etc) so that the server can innovate rapidly without revving the client (and also so that if, say, the source code leaks, none of the secret sauce does).

Overall, when I see this I think they are focused on the right issues, and I think their tool list looks pretty simple/elegant/general. I picture the server team constantly thinking - we have these client-side tools/APIs, how can we use them optimally? How can we get more out of them. That is where the secret sauce lives.

> but so little commentary on the core architecture.

The core architecture is not interesting? its an LLM tui, theres not much there to discuss architecturally. The code itself is the actual fascinating train wreck to look at.

The tools was mostly already known, no? (I wish they had a "present" tool which allowed to model to copy-paste from files/context/etc. showing the user some content without forcing it through the model)

Yeah in fact one thing claude is freaking great at is decompilation.

If you can download it client side you can likely place a copy in a folder and ask claude

‘decompile the app in this folder to answer further questions on how it works. As an an example first question explain what happens when a user does X’.

I do this with obscure video games where i want to a guide on how some mechanics work. Eg. https://pastes.io/jagged-all-69136 as a result of a session.

It can ruin some games but despite the possibility of hallucinations i find it waaay more reliable than random internet answers.

Works for apps too. Obfuscation doesn’t seem to stop it.

Whoa, when did they come out with JA3?

Why are "tools" for local IO interesting and not just the only way to do it? I can't really imagine a server architecture that gets to read your local files and present them without a fat client of some kind.

What is the naive implementation you're comparing against? Ssh access to the client machine?

It's early days and we don't fully understand LLM behavior to the extent that we can assume questions like this about agent design are resolved. For instance, is an agent smarter with Claude Code's tools or `exec_command` like Codex? And does that remain true for each subsequent model release?

It’s a distinction that IMHO likely doesn’t make much difference, at least for the mostly automated/non-interactive coding agent use case. What matters more is how well the post-training on synthetic harness traces works.

I found it a useful overview. My primary question about the client source was - is there any secret sauce in it? Based on this site, the answer is no, the client is quite simple/dumb, and all the secret sauce resides on the server/in the model.

I particularly valued the tool list. People in these comments are complaining about how bad the code is, but I found the client-side tools that the model actually uses to be pretty clean/general.

My takeaway was more that at a very basic level they know what they are doing - keep the client general, so that you can innovate on the server side without revving the client as much.

I can't read this article, but I live in a market that Waymo is just beginning to develop (St. Louis). I have seen two Waymos in our area in the last week. I am really, really looking forward to having it available, but I think it is several years away.

It will be useful for me and my wife, but it would have been transformative for my kids in the pre-driving teen years. We do so much ferrying them around, and to be able to put them in a safe car and get them where they are going would have made such a difference for us. I think it might be here in time to make a difference with my younger child.

For that use case, trust and safety are really, really important to me. That plan only works for me because I suspect the Waymo driver will be safer than I am. I think for companies developing self-driving cars, the safety reputation matters a lot. Right now Waymo is the only one I would consider, because their roll out has been so methodical and competent.

As a small business that started with a one-time/upgrade based pricing policy, and moved to a recurring policy, I don't think it is too late for tailwind to do so for future upgrades/improvements. I am saddened that they laid people off before trying. I understand doing that is a leap of faith/risk, but that is what you need to do.

The key thing they need to recognize is that some percentage of their customers are serious businesses that want them to continue developing/maintaining the software, and that these businesses will be supportive as long as the deal is the same for everyone (you can't ask them to pay out of the goodness of their hearts, as then they feel they will be taken advantage of by people who don't pay).

When we switched to a recurring pricing model, I thought it was going to be a disaster. In fact, I got an angry call from exactly one customer (who then remained a customer despite threatening to leave). I got subtly expressed approval/relief from many more.

The book "How to Sell at Margins Higher than Your Competitors" was helpful to me, and might be helpful here as well. The key is to realize that you want to sell to people who really value your product and will pay for it. You don't want to maximize volume, you want to maximize revenue x margin.

You already have an installed base of people who value your product enough to pay for it once, you just have to create a system that enables them to sustain the technology they value in order to get ongoing support/upgrades/fixes/etc. The people who are going to complain on hacker news about recurring pricing aren't the people you want as customers anyway.

If the majority of your customers don't value it that much, then you are pretty cooked. But you may as well find that out directly. If people really don't want to pay for the software, don't waste time creating it for them.

We made the switch about 20 years ago. Since that time, about 70% of our lifetime revenue has come from recurring payments. Had I not had the courage to make the switch, I would be writing now that the business has been an unsustainable mistake, but that would have been false.

>If the majority of your customers don't value it that much, then you are pretty cooked.

cries in gamedev

Sadly my options are to either sell a few thousand copies on pc and deal with complaints on how my game isn't an 80 hour long timesink, or go into mobile and employ all the dark patterns I hate about marketing.

I am surprised this hasn't gotten more attention. I feel like HN used to love nothing more than complaining about patent trolls. Anyway, this article suggestions an action through regulation.gov which, based on the content of the page, seemed worth doing to me.

Fast and easy to take action on that page.

Submitted. We’re still here!

"Nobody has any real values or beliefs anymore."

Have you heard of the concept called projection?

Cultural wars are strong as ever

"I feel like HN used to love nothing more than complaining about patent trolls."

Lamenting that everything is about AI, seems to be the thing of today.

AI has rather superceded intellectual property, since it's apparently fine to distil a derived work of everything on the planet if you have enough investor money.

Meta treatment vs aaron swartz made this crystal clear.

I think this is an interesting idea, but the details on how the incentives get aligned, and how graduated students support the university are unclear to me.

I am a fan of the idea that universities need a new pricing model that is correlated to career results. Like, maybe universities are only eligible for students to receive gov't loans in proportion to the increase in W2 income of their past students, etc.

But I am not convinced that the model in this article, as described, scales. It might be a model for attracting very high-value students and thriving. But in that case it might just be about selection effects, and not about delivering a value-add in education.

I also think that the article fails to recognize that one of the main tasks of major universities is research. Quite possibly the research and education functions should be separated, but I wish the article addressed some of these issues more explicitly.

Still, $100m towards and effort is interesting. Maybe it will evolve in an even more interesting direction.

The incentive structure of higher education is broken. My gift to the University of Austin is meant to change that—by tying its success to the real-world achievements of its students. By Jeff Yass.

I agree. I read an article a few months ago about how frequent MAID (medical assistance in dying) is in Canada. I am surprised that that has not led to larger scale studies about the dying process.

In this particular case, the press release notes "Scientifically, it's very difficult to interpret the data because the brain had suffered bleeding, seizures, swelling...". That does seem to limit how much can be generalized from this one case. A larger study of MAID patients would be more useful.

Edit: Maybe the issue is that the MAID itself would alter the brain state. That actually seems pretty plausible.

I think Steve Ballmer's quote was something like "Measure a software project's progress by increase in lines-of-code is like measuring an airplane project's progress by increase in weight."