Scott Ames-Messinger

Intercom: We will help you delight your users with your delighted Also @intercom We also hate our own users and make it impossible for them to to log into our own app.

Would not recommend @churnkey to anyone. I liked their product quite a bit, but their third party script broke our website and they've expressed no urgency about making sure this never happens again.

The @NotionHQ Agents just get worse. It asks for permission. Great -- I return to the chat. The dialog to approve is not in the chat. I click the desktop notice -- it takes me to my homepage. So thoroughly broken, I don't understand what type of feedback they want to get.

And the very useful history of AI chats (which is a main feature of ChatGPT and Claude) is just...missing.

And we're not even at the bizarre non-barrier between editing the settings and using the agent. They tried to blur this line so it makes me constantly second guess if I'm changing the agent's definition or just using it.

The new @NotionHQ agents are *so unbelievably terrible*, I don't understand how they were even beta quality. Me: Agent, go work on this document. Agent: I don't have access. Me: But I gave you access to the workspace: Agent: ¯\_(ツ)_/¯ And more permutations of access issues

@MDOTMTAHolly @jedweeks Fantastic! Thank you!

@jedweeks @scottmessinger Yes we all want tap to pay & are working towards it! In the meantime, having the ability to scan Charm Pass on Metro is an improvement so we advanced that effort. Our team did the majority of the work in house so we saved time & $$.

@jedweeks When I took the subway today, I noticed they have new QR code scanners for tickets. If they wer going to install something new, why that? Why doesn't the MTA have tap-to-pay like DC, NYC, etc?

I didn't know terminal apps could be great until I used @opencode It's incredible. Claude code feels like a glitchy MVP by comparison. But, in all fairness to @claudeai, opencode also eats memory like a hungry beast:

@joeychilson Also, it’s so weird the excitement CF has for this when you can’t use preview environments and their image service together. This is stuff that just works on vercel.

I wish this was true, but wrangler and bindings are holding Cloudflare back so much. I don't want to fundamentally change how I write apps just to be able to deploy on Cloudflare. Get rid of it completely.

I've generally been liking @CloudflareDev but given this 10 month old issue where preview envs break all a sites images (b/c they don't support the CF image service), it doesn't appears they're trying to seriously compete with Vercel. (github.com/cloudflare/wor…) /cc @eastdakota

@kylemathews @thruflo Ahh -- how it works now is great. I'll have to change the internals. I think claude got confused and I've also been confused.

@scottmessinger @thruflo you do have to create a stream first (though you can do an initial payload) and then keep appending to it. I have been thinking we should just let you append to a stream and it's created if not already there

@kylemathews @thruflo I made a Cloudflare Worker implementation of Durable Streams: github.com/commoncurricul… It passes the conformance test suite (thanks for doing that!) and also includes an optional pub/sub layer when a client needs to listen to 100s of streams.

@kylemathews @thruflo The touch/create might just have been how Claude internalized the protocol when it helped scaffold up the app. I'll have to go back and look. How it seems to work is you can either create a stream or just start appending and it will create one automatically. Is that right?

@scottmessinger @thruflo our use of PUT/POST is correct but does feel different than a standard rest API for sure. Not sure what you mean by touch/create?

@ariesrclark @kylemathews @thruflo Workers are request based developers.cloudflare.com/workers/platfo…

@ariesrclark @kylemathews @thruflo Nope :( You still have the calls to the worker. This is only really a problem if you have a lot of consumer and they're long polling. But, if you have lots of consumers with lots of activity, it's could be a ton of traffic and worker requests.

@scottmessinger @kylemathews @thruflo Huh, today I learn. community.cloudflare.com/t/avoid-worker… You can definitely still use the `caches.open` API, as exposed by Cloudflare developers.cloudflare.com/workers/runtim…. Would that be sufficient to keep costs at 18$/month?

@benjreinhart @elixirlang That's incredible! @josevalim Could we move channels between nodes when a deploy happens? We keep state in them and it would be incredibly handy if we didn't have to externalize that state

This isn't JSON, there's no database writes followed by database reads on the new node. This is one node sending its complex data structures as-is over the network to a new node in the cluster where it continues like nothing happened. The BEAM handles the clustering, network, serialization, etc. This is stdlib.

People often ask me what meaningful differences the BEAM runtime affords compared to other runtimes. Among many, built-in distribution comes to mind. @get_mocha we have thousands of long-lived, stateful agents running at any given time that need to survive deploys without any hiccups.

@ariesrclark @kylemathews @thruflo Nope. The worker always is triggered first. I would have loved to avoid the proxy. If you can find a way, let me know! It's highly annoying.

@scottmessinger @kylemathews @thruflo Can't you just use the Cache API or apply caching a page rule to prevent the worker hit?

@ariesrclark @kylemathews @thruflo Also, 1000% can be removed. It's entirely optional.

@ariesrclark @kylemathews @thruflo It's for the CDN. Since it's deployed on Cloudflare, you can't put the cloudflare CDN in front of their workers -- workers fire first. This means that it can get really expensive at scale for each worker request. The proxy gives you a way to put a CDN in front of the worker.

I think @s2_streamstore is pretty brilliant. @ElectricSQL 's durable streams has a nice API, but doesn't have a story around storage. I wish there was an s2 backend for durablestreams. Or, that the two companies would join up and make the durable stream company.

@kylemathews @thruflo Also, the semantics of PUT/POST are messing with my head -- PUT to create and POST to update breaks every convention I've been taught. Also, the touch/create style mechanics feel awkward to me.Conceptually, I'd prefer if there was no create and just updates or always create first

@kylemathews @thruflo Also, I just renamed the "subscribe" package to "estuary" (get it?). You said "multiplex" and I realized, oh, this should just be more generic instead of narrowly about sessions subscribing to streams.