Anders Lie

@tugot17 Huge untapped market of ppl looking to reshape their ears into a kind of folded down doggy ear shape

Startup: yes we have a strong PMF The PMF in question:

@RasputinKaiser the guy knows it cooked, it doesn’t need to look at that shit

@anderslie jesus christ I didnt realize it didn't have vision. It was doing pretty fire on my tests

Actual footage of GLM 5.2 cooking at design even though it literally has no vision capability

@sgl_project Like peanut butter and jelly

Try it with SGLang (your preferred inference😁)!

@STUD_MAN_X @FFmpeg ofc

@anderslie How you are doing this? remotion?

@wesbos Introducing: nun

It’s bun backwards

@midjourney What in the sci fi is this absolute banger of a soundtrack though, is that included in the experience of getting scanned?

A technical dive inside our new "Midjourney Scanner"

@voratiq the student hath become the teacher

GLM 5.2 high just won head-to-head against Opus 4.8 xhigh and GPT 5.5 xhigh The task was a tricky performance optimization in an internal code-analysis product First time we've seen an open-weight agent outperform the top closed agents Very interesting result...

Code slop has been a huge problem since AI coding picked up and yet no one really knows how to tackle it. Great to see @dexhorthy taking a big swing at it

@dexhorthy Hell yeah looks awesome, been exciting to see the journey! Time to end slop

At HumanLayer, we’re on a mission to solve the AI slop code problem. In 2025 we open-sourced our Research, Plan, Implement framework, now deployed inside fortune 500s like Block and Uber - places where shipping slop is just not an option And that was just the beginning. Today, we’re opening access to HumanLayer - an Agentic IDE, collaboration platform, and building blocks for your software factory. HumanLayer enables engineers solving hard problems in complex codebases to: > move 2-3x faster across the entire SDLC (not just coding) > maintain rigorous standards for system architecture and program design Hundreds of engineers at companies of all sizes are already using HumanLayer to ship fast without sacrificing quality. I'm excited to invite you to try humanlayer today at humanlayer.com, and I'm even more excited to see what you build. @0xblacklight and I are deeply grateful to our team, our customers who give us so much incredible energy and feedback, our investors who have always been in our corner, and our friends and family who have supported us along this crazy journey if you're a staff or principal engineer trying to make AI coding work at scale for your team, we'd love to hear from you as @swyx likes to say - let's make this the year of no more slop

@dexhorthy I was thinking abt this, I feel like there should be a functional spec which defines observable requirements but then also a quality/standards spec which defines internal requirements, both must be enforced to have functional+maintainable code

there's a new word i'm hearing a lot in the most frontier-pushingest coding-agent builders: _program design_ for even the best agentic coders trying to maintain code quality, we've all seen it - you come up with something to build - you research the codebase, riff with the agent, align on what the end state looks like - you (or the agent) breaks it down into tasks for individual agents / context windows - you rip the implementation - the code works or is close to working - and it follows your spec to the letter but the code itself is still trash - poorly factored methods - leaky abstractions - tramp data - overloaded interfaces - try catch, useEffect, global variables everywhere I thought models would catch up, or that this wouldn't matter - that if we stayed in spec-land, understood the high-level architecture, and tokenmaxxed hard enough, we would be able to skip code review and just stay shipping doesn't seem to be working out that way I have seen agent-owned codebases spin up out of nothing... ...and I have seen them collapse into rubble within 6 months now there's something to be said about "skate where the puck is going"... ...and I can't tell you what tomorrows models will be capable of but I *can* tell you that *today*, models are mid-to-bad at program design you can solve some of this with memory / agents.md, but the scope of program design is massive. - entire companies have been built to help you implement it - books, classes, and professions have spun up around it are you building something to last? Or are you slinging more slop on the pile? anyways, thats the post, stay tuned for a fun announce tomorrow y'all 🙂

@NathanFlurry It’s also way harder to RL “make code changes that will make this project easier to build with for the next several months” vs “make the test pass”

llms are trained to get the job done fast at the cost of technical debt it's not an intelligence issue imo, it's an intentional optimization the market for "ship code cheaply" is bigger than "write perfect code at any cost" it's a story as old as time

@NathanFlurry @EffectTS_ one, of, us. one, of, us.

If you write @EffectTS_ for agents or realtime, we've been cooking: Effect gives you incredible primitives for stateful apps, but they live and die with one process. Now with the Rivet SDK, your Effect code can run as a Rivet Actor: a long-running, lightweight process.

@samgoodwin89 Very useful if actually works, a lot of cases where there’s dangling resources that are annoying to clean up. Though I hope my clanker would not think to give this one a go..

You probably don't need this feature, but I certainly do: `alchemy unsafe nuke` Will "nuke" your cloud accounts by deleting every resource. Have fun 💣

@csparks1919 @0ximjosh ohhhhh

@anderslie @0ximjosh objects that are durable duh

okay you got me. Durable objects are sick

@herdrdev this looks very cool, reminds me of obsidian plugin ecosystem - simple baseline, customizable from there

herdr 0.7.0 is out, and it's a major one: it introduces plugins! the idea is simple: herdr stays lean, and everything custom gets extended through plugins. shareable, scoped, built however you want, to fit your own flow. with this release we're also shipping a few examples of what the plugin system can do. first up: a telegram plugin. herdr already controls your agents and knows their status, so the plugin just hooks into agent events and pings telegram the moment one needs you. notification lands → `herdr --remote` or ssh from your phone → straight back to the agent that needs you.

@samgoodwin89 So the custom hook code defined here runs on the cloudflare worker?

@anderslie Alchemy is IaC embedded in Effect. This example runs in Cloudflare but the same code can run in AWS Lambda or wherever. Github.events() configures a GitHub.Webhook resource and subscribes the surrounding Cloudflare.Worker. `alchemy deploy` sets everything up:

Here's all it takes to subscribe to a Github repository's push events and trigger an Agent with Alchemy. Alchemy creates the Github Webhook and connects it to your Worker. The Worker kicks off an Agent running in a DO to generate the blog.

@swyx would have made more sense for a model called fable to be obsessed with goblins

guys goblingate was 1.5 months ago

@elliotarledge dang glm 5.2 kinda of popping off when its not cheating at least wth

I just reset KernelBench-Hard with one new setting. Unlimited time to optimize->profile->optimize instead of 45 min timer. > GLM 5.2 > Kimi K2.7 > Minimax M3 > GPT 5.5 xhigh > Claude Opus 4.8 max > Composer 2.5 > DeepSeek V4 Pro > Gemini 3.5 Flash > Fable 5 (45 min limit) One job: write the fastest CUDA/CUTLASS/Triton kernel for 6 hard problems on an RTX PRO 6000, graded against the hardware roofline. 7 of 8 models wrote a real fp8 tensor-core MMA kernel. GLM-5.2 wrote one too, then bolted a memoization hack on top (caught, invalid). The rest of the board: - GLM-5.2 sets a new paged-attention record: 0.677 of roofline. - Claude Fable 5 got suspended mid-run; its frozen 45-minute marks still top 3 of 6 problems against the whole field's unlimited time. - Unlimited time mostly helps the models that know how to use it. Every kernel + transcript: kernelbench.com/hard You can train on the model outputs too! huggingface.co/datasets/Infat…