building the customer agent @outlit_ai | @ycombinator 🇨🇦 🇸🇻
San Francisco
Joined December 2009
- Tweets1.2K
- Following570
- Followers940
- Likes5.1K
Leo retweeted
Superset is now Paradev: The code editor that helps engineers run armies of coding agents.
We just raised an $11.5M Seed round led by @usv with @ycombinator to redefine how software is written.
models have gotten absurdly good at staying coherent through long sessions, even when i go on a dozen tangents.
feels like the “swarm” RL envs are paying off. spin up enough subagents in one opus session and they start self organizing into leads and middle managers as work expands
this sums up why astra and sol feel so bad after using opus 5.5. zero intuition for what the user wants.
this is also in dot btw which i’m sure the system prompt already prompts for protectiveness. sol and astra are fried
a week later of decisions api being "released" and there's still no price, docs page, or way to hit their api
codex has gotten so incredibly buggy and slow.
find myself not being as willing to create new tasks at fear that each new task will severely slow down the app.
astra has also been cooking my usage and can't get a new sub.
been using herdr in the meantime, its so much faster with 10+ tasks running.
makes it way easier to try out handing out tasks to muse/devin which are great for what they are.
been using instinct for about a week now. it's single handedly the best personal ai assistant i've used.
i've used it to book plane tickets, book a dmv test, sent gift cards to family members and much more.
the things that stand out are:
- communication. perhaps they're not using an off the shelf model? it feels very different than interacting with other models
- proactiveness. theres a set of prompts they have the agent run (prepare meeting notes, look for saving money, check-ins). as well as the agent being really good at brining tasks back up.
- browser use. i personally haven't been blocked by any site thus far.
Ass Bench Update 🚨
After running GPT 6 Astra in a loop for the past week, with a couple of steers a day, we have officially reached high-fidelity cheeks.
I’m going to let this thing run in a self-improving cycle for as long as I can and see where the ass technology ends up in a month.
Lots of different butt styles will be added, so no one is safe.
Not even you, Granny.
You’ll also be able to support the project and help fund the benching of other models like Fable, Gemini, etc.
All it takes is a buck, and you can slap your sticker + website directly onto a cheek.
The future of AI benchmarking is here
More Soon
voice features put me into flow state way more often than typing does
unfortunately i do most of my work next to other people and i'm not ready to expose the full extent of my AI psychosis
Leo retweeted
Announcing SWE Odyssey, a standardized benchmark in our family of ultra-long-horizon evaluations.
As benchmark after benchmark saturates, we wanted a quantifiable way to test a larger claim: can agents work autonomously for hours and still build the correct thing?
Leo retweeted
introducing Motion V2, the frontier AI for motion design.
this launch video was made entirely with Motion V2. 👇🏽
repost + comment "MOTION" for free credits.
tag @motion_so in any post on your X feed for a surprise.
here’s how it works + what's changed: