Jerry Zhou retweeted
It's been 3 years since AI exploded, but nobody's life has improved outside of Silicon Valley.
Tomo is for real humans, like my little brother Lucas
and 10,000 other humans taking a chance on themselves every single day.
I remember when @alexshander03 and @andrew_li03 told me they were thinking about building a company last year, incredible to see how far they’ve come 🥹
We’re launching @JudgmentLabs today and announcing $32M in funding.
As AI agents take on more of the work that creates economic value, they generate massive amounts of production data: the clearest record of how they behave with users, software, and the real world.
Judgment builds infrastructure for improving AI agents from production data.
release the delve files
benchmarking spatial reasoning is very important for eventually automating things like CAD, we hope to build open source tooling to help accomplish this!
Models can do hard math and write complex code.
But ask them to replicate a simple drawing step-by-step and they often break basic spatial constraints.
Preview results from our new benchmark: Printing Machines
Made with @jerryzhou and @fleet_ai
printingmachines.ai/blog/pri…
Common fleet W @nicoup @andrewthezhou
What does Applied Compute actually do? What does the data that Mercor actually sells look like?
In software land, evals and tool use capabilities are the last frontier to conquer before we can design models for enterprise use cases. To fuel this, we need engineering real world tasks to be deterministically completed and the ever rapid delivery of that data format at top shelf quality.
Fleet's description here is apt: "We aim to accelerate the shift towards the allocation economy, empowering humanity to transition from doing work to directing it—a capability previously reserved for the top ~1% of the global economy."
More in my piece on RLaaS and Human Data Markets seancai.com/philosophy/autom…
Jerry Zhou retweeted
added image support to our extension, which we plan to open source. stay tuned for more updates!
built a tool that lets you go to any website and edit it just by prompting. reskinned youtube and added a download mp3 button
made with @zephratic @jerryzhou
I need that creatine 🙏
we're giving out @cognition creatine (limited) 🦾🧠
quote tweet with (1) screenshot of your github commit graph (2) pic/vid of you lifting in the gym. we want to see both kinds of PRs!
fill out form below when done
Jerry Zhou retweeted
app builder update: you can use claude code now. watch it build me a superhuman inspired email app
more demos + release soon with @jerryzhou @jameszhou02
i've been working on a desktop app builder with my brother @jameszhou02. trying to get as much stuff to run client side and out of the box as possible has been a fun challenge
Fun blog on how we use some of Vercel’s features at Rox :)
Rox x @vercel = speed at scale ⚡
Together we’re shipping faster, scaling globally & powering thousands of AI agents in real time. Thrilled to be building the future of revenue tech side by side.
Big thanks to @rauchg, @perilanglois, @oynickj + team 💙
Blog by @jerryzhou and the Vercel team
vercel.com/blog/ai-agents-at…
Time to mod websites
built a tool that lets you go to any website and edit it just by prompting. reskinned youtube and added a download mp3 button
made with @zephratic @jerryzhou
📈📈📈
6 months, 25 million revenue agents & 3 trillion tokens later... Rox is now globally available 🌎
Just as coding agents 10x’d engineering, revenue agents 10x customer work.
With Rox, humans are evolving to orchestrators while agents manage the end-to-end customer lifecycle.
Even in Beta, Rox powered Global 2000 leaders in banking, hardware, construction, and sovereign AI, while serving dominant AI winners like @tryramp and @cognition.
Rox delivers ROI in 90 days and is built with the best.
Thank you to @OpenAI, @nvidia, @perplexity_ai, @awscloud, @vercel, @Snowflake, and @stripe for helping us scale.
Tech bros can now simulate the move to nyc to find a girl pipeline
I made a dating app simulator with @zephratic and @jerryzhou where you practice swiping. SF dating scene saved. Chopped people can use it for free.
We love @DevinAI and @cognition_labs 🤩
agent + tools? so yesterday. we’re in our group chat era now — agents talk to agents 😊.
w/our bffs (and beloved customers) — @cognition_labs — we’ve developed an agent-to-agent protocol that lets the rox agent swarm hit up @DevinAI to turn sales calls into instant demos.
it’s real, it’s live, and it’s already shipping demos 💪. lfg @russelljkaplan, @ScottWu46.
no manual prep. no generic decks. see it in action 👇
I built Orchestrator with @jameszhou02 , a proof of concept for how we envision the future of software engineering.
In the future, every engineer will manage swarms of AI engineers that execute their plans in parallel.
Orchestrator takes an input prompt and creates a plan that is executed by multiple AI IDEs (in this case Cursor) that work together in parallel.
At a high level Orchestrator takes a user prompt and creates an implementation plan with multiple tasks.
This plan then gets formatted as a graph, where sibling tasks are constrained to not have shared files to avoid file editing issues. These tasks get submitted to a task queue where waiting AI IDEs can pick them up to execute them.
When finished, each window is instructed to call the agent completion MCP server which informs Orchestrator that the task is done.
What we learned from building this:
- Hyper-specific prompts and execution plans with little to no ambiguity are needed, these should be generated via back and forth with humans in the loop for now. Execution should only begin when a detailed plan is developed.
- Vision and logs are incredibly important for application development. Right now there is only a very basic automatic feedback loop (using lint errors, compile errors, etc)
Overall this exercise was very insightful, with some better scaffolding/tooling these systems will become very good.
Full chess game raw recording: youtube.com/watch?v=oFi8CSom…