128GB of RAM and an M4 Max makes for a very solid development machine, and the build quality is a nice bonus.

But their tab complete situation is abysmal, and Supermaven got macrophaged by Cursor

https://aibenchy.com/compare/anthropic-claude-opus-4-6-mediu...

Who knew Anthropic was this far behind???

Every week it seems like we're getting closer.

Bonus: A high profile case might end people fixating on how long they can go without writing any code. Which makes about as much sense as a mechanic fixating on how long they go between snapped bolts without a torque wrench.

If I think something is worth the money, I typically don't need to actively decide to pause the subscription each time I use it.

But I checked and it's there... but in the UI web search can't be disabled (presumably to avoid another egg on face situation)

You can use this model for about 5 seconds and realize its reasoning is in a league well above any Qwen model, but instead people assume benchmarks that are openly getting used for training are still relevant.

From Sam Altman himself:

> We had this big GPU crunch. We could go make another giant model. We could go make that, and a lot of people would want to use it, and we would disappoint them. And so we said, let’s make a really smart, really useful model, but also let’s try to optimize for inference cost. And I think we did a great job with that.

4.5 scaled into a unified reasoning model would have been an incredible model. It beat GPT-5 on accuracy and hallucinations without reasoning (!)

It just wouldn't have worked for powering things like ChatGPT Go's rollout and loginless chatgpt.com, so they dropped it.

(And if you want, you could argue it's the compute crunch that didn't let them do both... but Anthropic had to make the same choices at the time and went in the other direction.)

If anything 4.5 being abandoned so they could sell India a $3 a month subscription was the first crack in The Box

You even open the article by linking the toy project where you used agents to "recreate Next in a week" and released with critical vulnerabilities.

That project you quoted is the one with that as its new description. Soon it'll just be [new thing] that happens to use the stars as social proof... in fact when I look again:

> The fastest repo in history to surpass 100K stars . Better Harness Tools that make real things done. Built in Rust using oh-my-codex.

They started a new project that justifies the same repo and scrapes a little credibility off of Claude Code. The intent is not an actual rewrite but to bolster what will be their own personal project trying to compete with OpenCode and co.

The grifter is already pasting references to WSJ articles about themselves in the Readme

"I have a popular repo, but the content will likely be removed and I won't have personally gained from the saga: how can I fix the part where I didn't profit?"

"Eureka! I'll remove the content preemptively, then come up with a backstory that justifies reusing the now empty repo for building the umpteenth coding harness! And I can even claim fuzzy ties to Claude Code!"

Hence the new description:

> The fastest repo in history to surpass 50K stars , reaching the milestone in just 2 hours after publication. Better Harness Tools, not merely storing the archive of leaked Claude Code but also make real things done. Now rewriting in Rust.

https://www.anthropic.com/news/detecting-and-preventing-dist...

Yes, some people left to their own devices would take twice as long to ship a product half as buggy only to find out the team that shipped early has taken a massive lead on distribution and now half the product needs to be reworked to catch up.

And some people left to their own devices will also ship a buggy mess way too early to a massive number of people and end up with zero traction or validation out of it, because the bugs weren't letting users properly experience the core experience.

So we've established no one is entirely right, no one is entirely wrong, it's ying/yang and really both sides should ideally exist in each developer in a dynamic balance that changes based on the situation.

-

But there's also a 3rd camp that's the intersection of these: You want to make products that are so good or so advanced *, that embracing the craft aspect of coding is inherent to actually achieving the goal.

That's a frontend where the actual product is well outside typical CRUD app forms + dashboard and you start getting into advanced WebGL work, or complex non-standard UI state that most LLMs start to choke on.

Or needing to do things quicker than the "default" (not even naive) approach allows for UX reasons. I ran into this using Needleman-Wunsch to identify UI elements on return visits to a site without an LLM request adding latency: to me that's the "crafty" part of engineering serving an actual user need. It's a completely different experience getting near instant feedback vs the default today of making another LLM request.

And it's this 3rd camp's feedback on LLM development that people in the 1st camp wrongly dismiss as being part the 2nd craft-maxxed group. For some usecases, slop is actually terminal.

Intentionally contrived example, but if you're building a Linear competitor and you vibecode a CRDT setup that works well enough, but has some core decisions that mean it'll never be fast enough to feel instant and frontend tricks are hiding that, but now users are moving faster than the data and creating conflicts with their own actions and...

You backed yourself into a wall that you don't discover until it's too late. It's only hypervigilance and strong taste/opinion at every layer of building that kind of product that works.

LLMs struggle with that kind of work right now and what's worrying is, the biggest flaw (a low floor in terms of output quality) doesn't seem to be improving. Opus 4.6 will still try to dynamically import random statements mid function. GPT 5.3 tried to satisfy a typechecker by writing a BFS across an untyped object instead of just updating the type definitions.

RL seems to be driving the floor lower actually as the failure modes become more and more unpredictable compared to even GPT 3.5 which would not even be "creative enough" to do some of these things. It feels like we need a bigger breakthrough than we've seen in the last 1-2 years to actually get to the point where it can do that "Type 3" work.

* good/advanced to enable product-led growth, not good/advanced for the sake of it

It started being proposed as a thought experiment "why should we care about the files if AI is going to do the edits", then as Opus got better and the hype built up, the rhetorical part of that dropped and now there are plenty of people who swear they don't write code at all anymore and don't see why anyone would.

I think we're in a feedback loop caused by the fact you can totally get away with not writing code anymore for some reasonably complex topics. But that doesn't account for the long term maintainability of the result, and it doesn't account for people who think they're not writing code, but are relying heavily on the fact we haven't fully magicked away the actual code. They're watching the agents like a hawk, doing small bits and pieces at a time, hitting stop when it starts thinking about the wrong thing, etc.

My worry is the market taking the wrong lesson out of the trends and prematurely trying to force the agent-first future well before the tools or the people are ready.