stillpointlab

Many CDNs do not support arbitrary Vary headers. Cloudfront, for example asks you to create a "Cache and Origin Request Policies" that gives you the ability to choose what headers are part of the request that gets sent to your origin. These will get added to the cache key, but it is a static list based on the request to the origin and not the response.

Akamai is another case where Vary is harmful, from their docs [1]:

    > As the content in response may be different for the same URL,  
    > Akamai  edge servers don't cache responses that include the Vary header, 
    > even if the content is cacheable by definition. The only exception is 
    > the case where the Vary header's value is Accept-Encoding and the
    > Content-Encoding header's value is br or gzip – edge servers cache 
    > such responses, applying the caching rules you defined in your property.

Cloudflare's docs do seem to indicate they support the Vary header as does Fastly. But one should read the docs of their CDN to find out the behavior. Do not assume Vary is supported.

https://techdocs.akamai.com/property-mgr/docs/rm-vary-header

I've spent time finally learning what tailscale is and how it works and I'm impressed. It's a rare thing in the technology world but I'm glad I finally took the time.

I literally just figured out how tsnet fits into the picture (an in-process Go based entire network stack that gets the process to act as a node in the tailnet) and so that helps me understand this (everything in tsnet excepting the control plane). It's very impressive that they can do this in a reliable way.

It took me some time to wrap my head around why WebMCP even exists. I was thinking that regular MCP + SSE/Websockets could cover almost all of the uses cases.

One interesting use case for WebMCP is cross-site activities that a browser agent might take. If the data-flow in question involves a heterogeneous set of websites.

This feels like a lifeline for incumbent web SaaS more than anything. Future apps can be architected in ways that do not require WebMCP but massive apps like Salesforce or Workday can't really abandon the decades of accidental business logic embedded in the Web UI flows that make up their project. While those same flows could be data-driven, slapping a WebMCP facade on top of them is just more practical. Instead of forcing incumbents to create an AI-native MCP where they are first class citizens, it is easier to let them sprinkle browser affordances throughout their human-centric front-end and offload the work to the agent.

But for my own part, since I am developing from scratch without that legacy need, I actually think relying on WebMCP might be an anti-pattern. It might be useful for the cross-site use case and it might be useful for reducing latency for purely UI activities (e.g. "filter this list" where all the data is already on the client), but in general my feeling is it is better to have a robust MCP interface for agents.

100% - the iPod wasn't just good enough, it was more than enough for almost everyone. The Zune was firmly in the diminishing returns category by the time it came out.

I think we might just end up in an argument about what constitutes "weird", or an argument about how good record keeping for killings over the last 6000 years have been. Although I think we could agree that Texas keeps better records about such things compared to villages in the dark ages. Even then, no record of why a violent act was committed likely would have mentioned "weird" and might only mention a dispute between neighbors, or an individual targeted by a mob. Then we might argue over per-capita or some other nit-picky thing.

At any rate, it appears the Witch Trials of 1400-1775 saw ~40k-60k people killed [1]. And in modern times, it is reported for example that ~20k people have been put to death in Tanzania for witchcraft [2].

While poking around Google and Wikipedia for this, it appears that in more stable societies the prevalence of these kinds of mass hysteria are low. However there appears to be sufficient evidence that during times of severe economic hardship, standard societal protections falter. In those circumstances, anyone deemed a burden or an omen of bad luck was at risk.

1. https://en.wikipedia.org/wiki/Witch_trials_in_the_early_mode... 2. https://en.wikipedia.org/wiki/Modern_witch_hunts#Tanzania

What Yegge describes in this post is something I have seen to a lesser extent in my own agent use.

But one specific thing I noticed, which his inline comic lampoons, is Fable (especially) explicitly stating "[User] has ruled", or "The rulings are in".

I was confused by this until I saw a post on X about how someone came to his agents in the morning after they ran all night, and the sub-agents had refused the requests from the orchestrating agents because they thought the decisions being made were not inline with the user. They thought the decisions were from the ochestrating agent.

This, as well as the governance things that Fable and Codex both often request, made me thing of provenance, especially related to decisions and authority. As we get deeper and deeper hierarchies of agents (as we likely will) this idea of authority, who has it and where does it come from, feels like it will be a key component to agental systems.

Morality advances when those who see they could, decide that they should not. In fact, one might see true morality as requiring a decline of reward.

If one abandons their principles as soon as a sufficient reward is offered, then they have no integrity.

So let your children see and learn what society rewards AND teach them integrity.

Of course there are many options, but the benefit of the SQLite approach is the existing tooling, support, etc. Like others have said, one could just write a SQL interface over a set of virtual tables derived from the actual elf (if one just wanted a CRUD-like interface to a binary).

All of the benefits of SQLite disappear once one diverges from the format in any way. At which point it would be better to ask: "what is the best first-principles implementation of this idea?", instead of "what is the minimal change to SQLite to achieve this specific narrow goal?"

It's interesting if you think of Solidarity in these terms.

There is a kind of solidarity of persistence under harsh natural conditions. And another kind of solidarity of persistence under laws.

So it might follow that the former group, those who feel above laws, also lack in a feeling of Solidarity. That would suggest a strong adherence to Individualism as a key component to their identity.

I'd give it some more time. My first interactions with Sol were not good. It cut a dangerous corner and I had to call it out.

But I continued to work with it and found that it was mostly my own style of interacting that needed to change. In a way it is similar to a new co-worker, they have their own personality and ways of working. Once I figured that out I have been able to get very good work out of Sol.

Sol seems to work better when you are clear, precise, direct and unambiguous. The model seems annoyed if things aren't spelled out. Not micro-managing, it seems to have a high bar for specific intent.

When I get Fable to write out specs for Sol, I tell Fable that Sol is a nit-picking literalist that is exceptional at instruction following. So far this description has lead Fable to generate specs that Sol implements at a high quality.

I think this is a narrow take, a bit like suggesting that women weren't historically discriminated against and putting forward Cleopatra as an example.

Surely, rich and noble eccentrics have existed in the past with several notable examples. But to point at exceptions and then suggests that this proves that no underlying trend is present seems like incorrect reasoning.

I was thinking that one could just write a loader that goes from SELF->ELF on load. You then pay the cost on startup. Probably not worth it for one-shot programs like ls/cat/etc, but for long running programs and daemons it might be workable.

My take, as someone who just read through the github issues conversation.

The model was outputting reasoning traces that were supposed to lead to tool calls. So the model might do something like:

    <think>
    I should use a tool
    </think>

    ... should make the tool call here

But a \n was slipping through from the last line of the reasoning trace so the parser was generating:

   <think>
   I should use a tool

   </think>

And that extra new line before the closing </think> would occasionally trigger the model to question itself with an "Actually ... " digression. In long running conversations this would end up looping because the "Actually ..." part would reason it should call a tool, then a new trailing \n would trigger an "Actually ..." and then it ends up in a loop.

Valid. There are many ways to do it.

But for enterprise there may be teams, each developing their own way to do it. Then there will be many different ways that it is done throughout the enterprise, which is hard re: governance. Better/easier to adhere to an industry standard which can be audited, especially for enterprises where that is a legal requirement.

That isn't a reason you should use it, just an explanation about why someone has to use it.

It is the automatic distribution and automatic update. The questions isn't "how does one download a text file to another persons computer?". It is "how does someone with a skill.md file on their computer discover that a new version of that file is available".

This isn't a "bottleneck" but rather a capability (or lack thereof). As you add more and more capabilities, especially ones relevant to enterprise situations like authentication, authorization, governance, etc. then MCP starts to pay off.

If you do not need those capabilities, then you do not need MCP. And then you shouldn't use it. But if you do need those capabilities then it might be worth using MCP rather than inventing your own way to do them.

The irony is, as I read the OP I thought it was very much in the style of McSweeneys and then there is this.

It is a pretty obvious humor device so I assume the ideas are independent. And the Socratic dialog framing adds another layer of humor since this is closer to a juxtaposition, lampooning both Swift's vapidity and Plato's writing style.

> Codex feels more like a version of Data from Star Trek

Great analogy for some reason. At fist I felt Codex Sol was a bit more cold. But now that I've worked with it for several weeks it has grown on me, even shown some personality. I appreciate that it is a bit more business-like, Fable is a bit too friendly sometimes when it ought to be focused on work. Codex can be a bit more nit-picky.

I agree with most of his other observations. I've already started to bin tasks based on which model I feel is best suited. In general, for well scoped and straight ahead tasks where banging out code is what I want I reach for Codex. For less specced tasks where I need a broader view and want the model to fill in more details I reach for Fable.

Both are great and they make a good team together.

I meant as a prerequisite to unlimited checkins.

Right now anyone can publish to public repos in an unlimited manner. They could choose to limit that and elevate unlimited to a paid co-pilot of codex bundled plan.

I wonder if Anthropic owning Bun will eventually lead to it being trained on in future versions of Claude.

Claude is already pretty handy with Typescript, but with all of the built in libs in Bun I wonder if there is even more to squeeze out.

They could do this in a way that lessens external pull requests.

For example, bolting co-pilot on to github, or a Codex in the web kind of thing that gives unlimited check ins.

It's like how Grok Heavy gives the user X premium or whatever. You charge for the tokens and give the unlimted premium access as a bonus. Basically, bundle it.

One of my pet peeves is people opening a parenthesis and never closing it. It's like my brain pushes the parenthetical into a stack and when I get to the end of the paragraph it throws an error.

I definitely overuse parenthesis and I haven't found a way to rid myself of the habit (as I managed to mostly rid myself of the habit of using semi-colons).

One of my go-to dad jokes is to see Ikea furniture at someone's house and ask "Is that a möstliglü [table|cabinet|etc]?"

The dark pattern I've noticed a few times is the exact match I type into Amazon search doesn't even return the item I entered but 2 or 3 entire pages of alternatives.

I know the item is there because Google returns it, either as a search response or in their "shopping" widget. I can click through Google to Amazon and see the item, sometimes even from the certified seller. But if I enter the exact search term in Amazon's search bar it doesn't show up in their own results.

So your saying the problem I am reporting is not a problem because it is obviously my responsibility to research the dozen available workarounds listed on internet forums, figure out which ones are not brand new footguns or genuine security risks, install sudo level programs from github that as the Warnings section on that repo confess "the daemon runs with super-user privileges and has full access to all private memory of all processes and sensitive user data;"

And then I scroll down and see the install instructions for my distro:

  > To install on Fedora:  
  >   
  > Orphaned for 6+ weeks, not available.  

It's funny I hear people slag npm all the time for developers just YOLO installing any dependency for the webservers they run.

But hey, just sudo install this package! What could go wrong? You only use your desktop to access your online banking, etc. Forget the trust mechanism of your distro and just live a little!

> how else do you expect it to work

How does it work on Windows and MacOS where I don't get the same freezes and random programs aren't killed by the OS? Because I'd expect it to work like that.

There is this false dichotomy, a black and white kind of thinking on display.

> And I really doubt you mean to say that Windows is _more RAM efficient_ than Linux

What I'm saying ought to be clear to anyone willing to listen. My system should not get into an unrecoverable freezing state when I am doing routine tasks on a fresh install of an OS. And the community that is supposed to champion this OS should not then shift the blame to me if I complain about it.

I mean, I'm to blame at the moment as well. I'm putting up with it because I don't want to use Windows, mostly because I'm doing a lot of agent dev and I saw them struggle in a Windows environment. And there is a little spite driving me since Windows 11 is full of ads and I felt like it was pushed on me when I wanted to stay on Windows 10.

The vast majority of my computer use has been on MacOS for the last 15+ years, since a Macbook Pro is the standard issue dev environment at every company I have worked at. I had a Windows desktop entirely for gaming that I've repurposed to be a Linux dev box. I genuinely thought all of the hype about Linux desktop that I have seen lately was an indication that it was finally ready for general use.

Now that I am vigilant watching RAM I haven't had the freezes. And I refuse to let the OS roll the dice on what program it wants to kill in its last ditch attempt to save itself from a problem it shouldn't have in the first place. And since Firefox seems to demand 800MB-1GB per tab process, it is a reliable scheme to just close heavy tabs when I see memory dip into the danger zone.

I just wish RAM wasn't so expensive right now, since I'd buy as much as I could just to avoid this pain.

This gripe (that I literally crashed out on recently after maining Fedora for the first time) is top of my list for annoyance/frustration with Linux as a desktop.

I now have btop permanently running in a Terminal and I habitually Alt-Tab to it to watch my swap. As soon as my free memory gets less than 1 Gb I kill any Firefox tabs (the memory black hole).

I know this won't be popular, but Linux people have some kind of Stockholm Syndrome on this. They tell you all of the obscure configuration files you have to edit, the additional programs you install. They insist that just having the OS kill a random program to save the DE is a good plan. It's the typical "your holding it wrong" mantra.

I run both Windows and MacOS and have done so for decades. I cannot remember the last time I had a hard freeze like I get on Fedora. On the exact same box running Windows I was able to have VS Code, Chrome with 15+ tabs, terminals, docker, etc. all running with zero issues. On Fedora I have a handful of Terminals and Firefox with less than ten tabs and nothing else running and I have to watch btop like a hawk.

The excuses people make are mind bending, and the shifting of blame from the system onto the user is shameful.

I'm not claiming an eval on what is or isn't possible. Rather I am stating my satisfaction with what I actually used and the actual result I obtained.

What I can say is that over the course of ~1 week I was able to review, plan, implement, test and release a significant change to a production system using Code Sol max. Any other claim about how any other model might have completed the same task is outside of my experience.

I like to see this. I still prefer Fable (marginally) but my last big task was 100% Codex using Sol max (re-sizing my AWS infrastructure using CDK) and it did a very good job. No complaints, I could use this model happily to do what I need to get done.

If this nudges Anthropic to give me more Fable usage, that's even better.

Sounds like a business opportunity.