wireless

Hiring - New Grad, $300K USD all cash I'm on a well-funded stealth team working on an AI consumer app that's grown to 50K DAU in the last year. It's an extremely fast growing space with lots of interesting challenges. From a technical POV, the product is similar to a vertical-specific Discord i.e. multiplayer real time, client-side sync engine, low latency moderation systems, etc. We're looking to hire an engineering team member that can work across the stack. You'd be joining a very strong team of <5 people (ex-FANG/Quant Trading) and you'll have a lot of autonomy to drive initiatives end-to-end. - Work remote from anywhere - Base comp: 150K USD - Bonus: Minimum 150K guaranteed first year. Determined by performance, scales beyond base comp for strong performers. - Years of experience: Don't care, we're interested in your ability and work ethic Please DM me if you're interested!

@izadoesdev Does the new clickhouse JSON type fix this? I remember checking a while ago and Posthog was still on the old one.

one of the reasons posthog is so slow is because of the way they structure their database by default, everything is a "property", the entire event structure is a JSON format, so even the basic information ends up in a way that isn't very optimized in clickhouse the only reason for this is because they allow you to collect data in any shape you want, but realistically not everyone uses this, and it's not even necessary (seperate database tables for basic page views / events vs custom events) in databuddy the opposite is true, we have a rigid shape for basic events because those never change, and it allows us to show your data way faster, and more efficiently custom events live in their own table where it has it's own JSON you can add properties to

@maria_rcks @theo huh, think I would actually prefer this UI to opencode's

Since we all know that terminals are made for complex UIs... I decided to make T1Code (1T, because a terminal is all you need). I know @theo really likes this kind of complex UI right on the terminal... so lets hope he likes it!

@lostbutlucky pretty easy these days via TTPS. dont need employment

Anyone know how Geohot got his Hong Kong residency? And has anyone done Dubai?

@bourneshao @paper

@wireless_dev which tool? been bouncing between a few for UI mockups and most of them still feel like glorified templates

holy shit, just tried this out. unreal. all the other ai design tools are toys in comparison

@samhogan I think Zed is working on DeltaDB for this use case.

What if a codebase was actually stored in Postgres and agents directly modified files by reading/writing to the DB? Code velocity has increased 3-5x. This will undoubtedly continue. PR review has already become a bottleneck for high output teams. Codebase checked-out on filesystem seems like a terrible primitive when you have 10-100-1000 agents writing code. Code is now high velocity data and should be modeled at such. Bare minimum, we need write-level atomicity and better coordination across agents, better synchronization primitives for subscribing to codebase state changes and real-time time file-level code lint/fmt/review. The current ~20 year old paradigm of git checkout/branch/push/pr/review/rebase ended Jan 2026. We need an entirely new foundational system for writing code if we’re really going to keep pace with scale laws.

@boneGPT @nikitabier If you need help with this feel free to DM me.

FYI if your accountant doesn't already know about QBI optimization and/or 2/7ths rule (where the 28.5% comes from), you're probably better off switching to one that does. There are a bunch of other caveats (e.g. SSTBs vs non-SSTBs). It's actually a good litmus test for screening firms. Realized this after trying to explain everything to my previous CPA and ended up rage quitting. The accountants that know about these things tend to be savvy in a bunch of other ways.

my business is doing 15-20k a month and i can't qualify for a mortgage i need to pay myself a higher salary for TWO WHOLE YEARS before i can qualify for one, tripling my tax burden might be smarter to keep renting

@samhogan think zed is doing something like this with deltadb

@_ashleypeacock Durable objects are always hot during IO, they can hold on to non serializable objects e.g. Promises

With Workers being billed for CPU usage, and not wall time, I wonder why Durable Objects bill wall time still? Considering most of the time you’re waiting for the response from the LLM, the cost for AI agents on Cloudflare would be cheaper than anything else by a wide margin I suspect, if this were changed

@GregKara6 @BenjDicken yea, pretty much

I don’t think you understand how SpacetimeDB works You are evaluating SpacetimeDB through the lens of traditional database benchmarking, but SpacetimeDB isn’t trying to be a better Postgres it’s eliminating the entire layer between client and database. The “pipelining” and “Rust client” aren’t benchmark tricks they’re the product. SpacetimeDB runs your server logic inside the database as WASM modules. Clients connect directly via WebSocket. There is no separate app server, no ORM, no HTTP round-trips. The Rust/TS/C# clients are auto-generated typed SDKs So when you say “why does SpacetimeDB get a custom Rust client while Postgres gets a slow JS client with an ORM over HTTP” that’s not an unfair benchmark setup, that’s the entire architectural thesis. You’re seeing the difference between “database that an app server talks to” vs “database that is the server.” Comparing it to a 3-node HA Postgres cluster is like benchmarking a monolith against microservices on a latency-sensitive workload and calling it unfair that the monolith doesn’t have network hops. The whole point is removing those hops. The real question isn’t “is this a fair DB-vs-DB benchmark” it’s “does this architecture deliver faster end-to-end for real-time apps?” And the answer is absolutely.

Is spacetime cool tech? Yes Are these good benchmarks? No I've done a lot of database benchmarking. It's fun. I like benchmarks! But these were poorly conducted. - So many missing details. What cache warming was done? What region was each database and client in? What were the underlying instance types? Are we benchmarking databases or network latencies? - Anyone create a narrow benchmark that makes a particular database look good. It's better to use widely-used standards (tpcc, sysbench workloads, etc) - AFAICT the spacetime requests are heavily pipelined vs the Postgres options. - Why does spacetime get a custom rust client (rust is fast!) and all the others have to run slow js clients with an ORM middleman over http/rpc? Ultimately it's a big case of apples-to-oranges. Again, spacetime seems like cool tech! But comparing it to a 3-node HA semi-sync pg cluster is... an interesting choice.

@chronark @Cloudflare iirc it's even worse when you have enterprise SSO turned on, you need to Continue with Google twice lmao

hey @Cloudflare can you please fix your logins, it takes 30s to go from cloudflare.com to actually being in the correct workspace 1. go to cloudflare.com 2. click deny cookies 3. click "Log in" 4. cloudflare loading spinner 5. click "Continue with Google" 6. captcha loading spinner 7. click captcha 8. captcha loading spinner 9. cloudflare loading spinner 10. 2FA appears 11. fill 2FA with 1password 12. cloudflare loading spinner 13. account loading skeleton 14. click account 15. cloudflare loading spinner 16. click on whatever product I need 17. cloudflare loading spinner @ritakozlov please

@samgoodwin89 @bouguereau_stan Experienced something similar e.g. stand up RDS cluster, now you want to stand up a DB within that cluster, but doing so requires you to connect to RDS and therefore go over tunnel

@bouguereau_stan I don’t totally know what you mean. Can you share more?

Here's all it takes to create a S3 Bucket, SQS Queue and forward Bucket Notifications to the Queue with `Stream.tapSink`. Alchemy-effect creates all the Resources, IAM Policies and Environment variables automatically. Plan is to have a release ready for usage this week.

@boristane @CloudflareDev @dok2001 @ritakozlov @bjyule @irvinebroque @elithrar @baselimehq congrats and looking forward to what's next!

friday was my last day at @cloudflaredev two years ago @dok2001 @ritakozlov @bjyule @irvinebroque @elithrar took a bet on @baselimehq what followed was the most rewarding chapter of my career, shipping with a team that genuinely raised my bar every day grateful doesn't cover it

@cis_female @0xdoug @balajis has layer looping ever worked at scale? think that alone would be a bigger deal than the entire rest of what they're working on.

@0xdoug @balajis looks like they fit the 8B on 10 dies. In any case if you went hard on densification + maybe stuff like universal transformer maybe you could make it work. Nobody's really been optimizing for perf/parameter hard, seems plausible there's a lot to squeeze

Soon we’ll combine (a) the speed of execution seen in Taalas with (b) the duration of execution seen via METR. And get days of work done in seconds.

@AlpinDale @yacineMTB wtf im renting 4090s for double that

@yacineMTB IDK I rent 5090s for about 20 cents an hour.

bruh fr?

@lefthanddraft @teortaxesTex It was free, that's pretty much it

Is there any eval that captures why 4o is an outlier? Despite its reputation, Anthropic's testing found it scored less than Opus 4 on sycophancy, user delusion, spirituality and bizarre behavior ...

@teortaxesTex @lateinteraction Maybe comparing base model vs the best open source non-reasoning models

@lateinteraction > it already beats the best open-source models before any fine-tuning or RLHF. Just pretraining I"m pretty sure it means that the Avocado base model beats Instruct models. We've heard the same about Behemoth btw

> it already beats the best open-source models before any fine-tuning or RLHF. Just pretraining bearish on Avocado's potential if true The kinds of skills modern models show, 100+ step agentic traces etc, should not be possible with good faith pretraining

wow, super bullish on this. maybe this is jumping the gun a bit but one thing I've always wanted was a unified observability + product analytics tool. like I want my DAU, MAU, funnels, conversion tracking, etc. derived from my obs data, instead of needing to integrate a separate SDK like amplitude/mixpanel (which are often way less rich than OTel data anyways).

Maple is now almost here! You can now waitlist Maple and I will start inviting people into the private beta sometime next week. Open Source and self hostable. maple.dev

@THEM2DT @im_roy_lee @ishaansehgal way more than 200

@im_roy_lee @ishaansehgal no way 20k, maybe 200 if its really solo

The first $100K/month solo founder with zero employees is coming. Not next year. Soon. Here's what this means: • No payroll, just API costs • No management, just monitoring • No hiring delays, instant scaling • No office politics, pure execution Entire businesses running on autopilot. The one-person empire isn't a dream anymore.