will

I’ve been obsessed with AI gaming for over a year now and have loved seeing all of the new ideas and games pop up! Thought I’d break down the way I see the space and try to categorize some of it. 🧵⬇️

@waynesutton I believe convex can scale but MAN is it expensive. $2.2 per function call and the fact that every query / mutation requires a function call on top of the already very expensive ingress/egress is painful. We spend $2.5k/month on supabase and just built a v2 of our app on convex. With only 5 users in our alpha testing we’re already hitting bills. I’m sure some of that just needs more optimization on our end but even once we’ve fully optimized it feels like convex is still going to be somewhere 3-5x as much as what we’re currently paying. Devex is great but not sure it’s worth that much. Would love to chat with someone on your team to help us get those costs down.

What a wild day in tech. Meanwhile the convex team is helping customers apps that scale and save less on their convex bill. Now whenever someone ask can convex handle scale, I’m going to send them this post because 99% of those asking are nowhere near the numbers of clawhub. Moral of the story, more contributing, less forking…

@haydenbleasel thanks for all the hard work!! have loved your work. who is taking over json render :o

Today is my last day at Vercel. I've had an absolute blast over the last year. I got to work on a lot of my own projects that helped form the basis for our AI Cloud, helped other teams ship their great ideas, and worked alongside some of the most talented people I've ever met. I'll have news on my next adventure for you on Tuesday but for now, wanted to reflect on my time here and the projects I worked on. Read on if you're interested ↓

How are folks generally handling counting tokens with the @aisdk when users abort?

@PaulRBerg totally agree. because sometimes my last copy didn’t actually work

This is the most annoying thing in Claude Code. Hiding raw text when you paste more than 4 lines. Terrible UX decision.

@nikitabier i can think of a few reasons why it might be especially slow for x lol

iOS developers: How long is App Review taking for everyone these days? It is now taking longer to get our app approved than it is to build the actual features.

@aisdk we really need abort signals to be compatible with resumable streams in v7! Thank you all for your hard work 🙏🏼

@emollick could apps use faceid api to verify human before post?

I know I go on about this, but comments to all of my posts, both here and on LinkedIn, are no longer worth reading at all due to AI bots. That was not the case a few months ago. (Or rather, bad/crypto comments were obvious, but now it is only meaning-shaped attention vampires)

@brycent We did. the sync is so good and way better to work with an agentic repo but auth is awful and support is much worse. @convex @supabase

Debating moving a project from Supabase to Convex Anyone whose done this, can you vouch?

Say it with me: every inference provider should return the per-request cost in the API response. Pls rt.

@rseroter @Google @rseroter do you any mechanisms to balance out highly visible and tangible roles vs quieter but really important roles?

Inside @google, we have a system for sending small bonuses to peers that helped us out. It's used often, and builds a culture of gratitude. We added an AI tool that scans your chats, emails, whatever and generates a report that shows who helped you the most lately. So handy.

@Sherveen @jeffmurphyyy agree I feel he is so cringe these days

@jeffmurphyyy If he wasn't consistently next-level cringe, I'd give him the benefit of the doubt, but...

To be clear -- (1) Garry should be embarrassed for tweeting this. (2) If it's true, that CTO should be fired immediately. (3) Whenever I think I've finally hit the bottom of "VCs are mostly stupid + lucky (other than my friends)," one of them manages to find a new bottom.

@RhysSullivan as someone who has tried a lot of things with mcp/cli, my life just got a lot simpler with cli and everything works. mcp had too much pain

the overindexing on CLIs is kind of insane to me it's building a primitive that's not portable, properly discoverable, has no good approval flow DCR / CIMD to APIs would go so much further but CLI is just the current hype thing

@trq212 I turned it off right away because I found it too black-boxy and preferred to have my own instructions for managing memory

now that it's been 2 weeks, what have you liked most about auto-memory? what's not working? if you have specific stories of when it went well or poorly, we would love to hear!

@trq212 for an infinitely running conversation (assume ai game master playing a game of d&d), how would you think about caching? With implicit caching not being guaranteed between turns anyways, do you try to append only and preserve cache until compaction, or do you compact / rejigger on every turn and only hope to cache follow up steps? Or do would you do some hybrid? Curious what your intuition is here - I feel like it changes a lot when its one long running conversation with a lot of narrative context to manage vs a bunch of smaller conversations around specific tasks

@tobi As an avg developer who is scared of anything that actually resembles real ML but uses a lot of LLMs, would I be able to figure out autoresearch to improve some web app repo?

Autoresearch works even better for optimizing any piece of software. make an auto folder, add program.md and bench script, make a branch and let it rip.

Anyone build a library for the @aisdk that helps defend against common problems with popular models? For example, gemini outputting <ctr46> or a model that bugs out and starts repeating the same tokens over and over? Looking for prebuilt defensive code to handle model quirks

this claude ai review pricing makes no sense compared to the normal claude code subscription. probably an early warning sign the sub prices are over soon huh

@trq212 phew okay thank god!! appreciate the reply

@consolelogwill yeah sorry regression, will be fixed in the next version

@trq212 did you guys change new lines from shift+enter -> option+enter? My claude code all of a sudden changed today and it's throwing me off so bad

There really need to be agent first ways to search: flights, airbnbs, hotels, etc.