beecasthurlbow

They also halved API prices with GPT-6, so I think it’ll net out pretty close for sol and Luna, but much less to be used for Astra.

> I don't understand how this helps. CNAMES have TTLs like A records and they eventually have to terminate at an A record somewhere so why pay for an extra hop?

I assume the customer controls the domain DNS records here rather than the hosting provider.

CNAME record: hosting provider can change underlying IP freely.

A record: hosting provider must get in touch with customer to change DNS.

So do you still have humans doing code review? What did “sign off on code” look like in practice? (We’ve been thinking about this a lot too)

What plans did you put in place? What did you do to ensure knowledge transfer?

It seems like you're saying "This is a platform to iterate on your agents and make them better over time", but the headline of your site does not communicate that at all.

Your website is full of AI-isms, which makes it hard to parse what you're doing. (3 punchy sentence headline, "X is not Y." , "never able to claim").

Sounds like you have a cool product, try writing at least the front page of your site yourself and it'll be much much more engaging.

What does "changing the specialist" mean? changing the model? The prompt?

I’ll try it thanks! I just got good mileage out of “don’t use ‘X, never Y’ format” on a document too, but I think linear deductive flow Covers it

> I expect the gruesome writing style of LLMs to be fixed relatively soon, seeing how it can already be prevented with just a few lines of stylistic instructions

What instructions do you use to prevent the gruesome writing style? I have… not found this to be the case.

Agreed. Throwing it back into an AI tool to condense it helps, but def hard to read.

I’ve been using both, but I find 5.6 sol is faster in terms of wallclock time and far less chatty (on my small projects in non-legacy code bases).

It “feels” like Sol gets off track more easily than Fable (it’s hard for Sol to ignore a stray remark). My theory is this comes from smaller context window - Codex uses a 350K token context window (that you can’t change), even though model can handle more…

Can you make it clear on your homepage what you have to do to use this? I.e., do I add a library to a codebase? Do I add opentelemetry to every request? Does it read my logs?

I went and tried to skim your docs here (https://workers.io/docs/quickstart) and I still don't get it. Looks like there's something around reading logs?

What do you mean? Could you expand on your answer?

His explanation seems pretty reasonable, no?

> He pointed out that the Connectors page featured OpenAI and Anthropic, two companies that hadn’t contributed to the WordPress project and nobody had complained about their inclusion. He said Google had contributed in the past but had stopped. > “… how ridiculous is it attacking Akismet, and Automattic, and blocking the thing the person who is our [release] lead asked for,” Mullenweg wrote.

That may be true for this specific use case, but the protection on bigger services could be easily rolled back. E.g., law enforcement was able to ask Google for "tell me everyone who was in this specific area during this time" [0] and is still able to ask "give me everyone who searched for a specific term" [1].

[0] https://www.forbes.com/sites/larsdaniel/2024/10/08/google-to... [1] https://reason.com/volokh/2025/12/16/are-there-fourth-amendm...

They asked Google for “who searched for this address or this person’s name” and got back a list of people. That’s chilling (vs - eg - “what’s this specific person’s search history?”).

Makes you think about this capability if the government decided to crack down on specific types of speech.

Google shouldn’t store this information in the first place.

If new apps are automatically surfaced, how do you know you’re getting the authentic app? Eg “Zilow” vs “Zillow” or something… is this the new typosquatting?

    ChatGPT can also suggest apps when they’re relevant to the conversation. For example, if you’re talking about buying a new home, ChatGPT can surface the Zillow app as a suggestion so you can browse listings that match your budget on an interactive map right inside ChatGPT.

    […]

    As more developers build with the Apps SDK, the list of what’s possible will keep growing. Because these apps appear naturally in your chats, you’ll find them when they’re most likely to be useful.

Looks really cool! I had slight trouble understanding whether the repo is the complete codebase or if it connects to a separate backend.

Does one host the server or is it connecting to Airweave backend? Put another way: how does Airweave make $$ and where does data stay at rest (if it stays anywhere)?

Doctorow also focuses more on the decision NOT to expand flocks (which comes late in WSJj. Really easy to say “it’s just supply and demand” while obfuscating that you’re limiting supply.

An easy (ish) option here is to use autosquashing [1], which lets you create individual commits (saving your work - yay!) and then eventually clean em up into a single commit!

Eg

    git commit -am “Starting work on this important feature”
    
    # make some changes
    git add . && git commit —-squash “I made a change” HEAD

Then once you’re all done, you can do an auto squash interactive rebase and combine them all into your original change commit.

You can also use `git reset —-soft $BRANCH_OR_COMITTISH` to go back to an earlier commit but leave all changes (except maybe new files? Sigh) staged.

You also might check out `git reflog` to find commits you might’ve orphaned.

[1] https://thoughtbot.com/blog/autosquashing-git-commits

16 participants and 20 controls is small enough that you might miss a subset of people that do have reactions to gluten. My associate has huge migraines when he eats gluten, but does not have celiac disease.

Looking forward to more research here (and at least they said “casts doubt” rather than “disproven”)