ec109685

What do you see as the main advantages of gvisor instead of working within a lightweight VM with full native performance?

As the container escapes with K8s shows, it is super tricky to get isolation right. E.g. what happens if a file you think is safe to write to is suddenly is replaced by one that isn’t.

I was expecting something much more rigorous than a few trials resetting the controller at random points in its update process.

> to me this shows the importance of human in the loop

They can’t decrypt the thinking traces.

Fable is still the best there is. Sol close second but I find it gets way to stuck on details.

Also Opus 5 is fine if your codebase is simple.

I spent $800 in a few hours when my sub maxed out because I was trying to get something done and had a long car ride to let it churn.

Their api pricing is absurdly expensive.

> The bug is a data race with tight timing constraints. It is unlikely to occur in common use. The developers have never been able to reproduce the bug organically and had to add special testing logic to SQLite that deliberately triggers the circumstances of the the bug in order to verify that the issue has been fixed.

I read that and it feels like some theoretical thing versus causing impactful outages for workloads shaped like tailscale’s.

Tailscale isn't mentioned in the SQLite changelog.

I agree it's tricky, but Tailscale had downtime by this bug and the SQLite Changelog downplayed it.

While technically true as written, it seems to downplay the significance:

> The bug is a data race with tight timing constraints. It is unlikely to occur in common use.

A large customer did experience this corruption, so it's important for people with tailscale's setup update immediately.

> The developers have never been able to reproduce the bug organically and had to add special testing logic to SQLite that deliberately triggers the circumstances of the the bug in order to verify that the issue has been fixed.

I try to use all the intelligence I can, which fable, opus and then second tier models.

I am not sure why you wouldn't want to use the SOTA models unless speed is a concern. Otherwise you are leaving quality on the table.

Yeah I wanted to do that at previous company. Got talked out of it, but it's nice have all those ips available.

One honesty note before my comment (yuck), it's super frustrating to read an LLM produced blog that could be 1/3 the size. Why are you talking about MCP's when comparing OpenCode and Claude Code when they both support that technology?

The only interesting thing is whether OpenCode is more effective at writing code with it's reduced system prompt.

Caching doesn’t work the way the bug reporter implies. Caches are shared (at least across the enterprise), but its key is always a function of the input before it.

We achieved significant savings simply by moving everything that varies across individuals out of the system prompt so every session starts from a cache point.

For example you never want your system prompt to start with the time that the session started. Move that to the first user message if needed.

Not true at all. I hate reading my own produced AI writing.

The issue with any AI writing is that it all sounds the same.

Once that stops being true, maybe it will be acceptable. But until then, you are left with repetitive crap. That you must wade through. Not good.

Missing gradual rollout of feature flag changes themselves. Yes, you can do percentage based rollouts for individual features but still should have ability to canary all changes before they cause an insta-sev.

Supply and demand always balance out. There is no way manufacturers aren’t going to compete away these inflated margins, as long as they feel like this demand is sustainable.

The issue is they become pay to win, which just isn’t as much fun.

There is no way a port this massive will have human code reviews.

If this succeeds, there is no stopping AI given it will have crossed the rubicon of human bottlenecks.

> “Here’s a million dollars to implement WebRTC for the fourth time”

“Hell no”

> “Umm…”

It actually flying: https://www.instagram.com/reels/DXrvm_LAAR8/

He would pull them away from co-pilot and the unlimited spigot of money that agentic coding brings, which is contrary to the best interests of Microsfot.

Wonder what they do for their token cache if they swap mid-session like that.

Did they nerf the model or was it changes to Claude code? I agree it got frustrating.

Great example as to why people are yearning for CSS in TypeScript. Something as simple as if() only works in Chrome and there's not a good shim story for CSS versus a more complete language, so you end up with this:

> The problem: CSS can compute a number – 0 for visible and 1 for hidden – but you can’t directly use that number to set visibility. There is a new feature coming to CSS that solves this: if(), but right now it only just shipped in Chrome.

> So I used a trick called type grinding. You create a paused animation that toggles visibility between visible and hidden. Then you set the animation-delay based on the computed value to determine which keyframe is used:

  animation: cull-toggle 1s step-end paused;
  animation-delay: calc(var(--cull-outside) \* -0.5s);

  @keyframes cull-toggle {
    0%, 49.9% { visibility: visible; }
    50%, 100% { visibility: hidden; }
  }

> A negative animation delay on a paused animation jumps to that point in the timeline. So a delay of 0s lands in the visible range, and -0.5s lands in the hidden range. It’s a hack, but a functional one. When CSS if() gets wider support, we can replace this with a clean conditional. ```

I think Anthropic just highly RL’s their model to work best with it’s Claude Code’s particular ways of going about things.

All the background capability Claude code now has makes things way more complex and I saw a meaningful improvement with 4.6 versus 4.5, so imagine other harnesses will take time to catch up.

Me too. And love pulp fiction. Just used Mr. Wolf to reference a situation at work.

Defense in depth.

Any large organization is going to have some terrible employees.

EC2 spot instance, Claude code, Prompt from Panic, Eternal Terminal.