Research lead for https://teapawt.net/t.co/Ax69nGsKRw @ AWS. PhD from @Berkeley_EECS. Co-organizer https://teapawt.net/t.co/mV8bqpqvV7. Views mine.
Joined July 2012
- Tweets1.1K
- Following301
- Followers3K
- Likes1.3K
Shadaj Laddad retweeted
Super neat project. Congrats Shadaj and team on the release! Clever way to re architect tool orchestration/execution for agents; the de facto paradigm of forcing agents to spawn background or foreground jobs (and rely on os primitives) seems super silly in hindsight
So excited to introduce Infinity, an agent runtime in Rust for extreme concurrency!
Infinity brings a new architecture that treats agent concurrency as a scheduling problem, tool call primitives for asynchronous events, and native support for serverless.
github.com/hydro-project/inf…
Shadaj Laddad retweeted
“light enough to fit 75k agents in the memory of a Raspberry Pi”
So excited to introduce Infinity, an agent runtime in Rust for extreme concurrency!
Infinity brings a new architecture that treats agent concurrency as a scheduling problem, tool call primitives for asynchronous events, and native support for serverless.
github.com/hydro-project/inf…
Shadaj Laddad retweeted
Want a glimpse of the future of agent runtimes? Check out some amazing innovations from Shadaj & team:
So excited to introduce Infinity, an agent runtime in Rust for extreme concurrency!
Infinity brings a new architecture that treats agent concurrency as a scheduling problem, tool call primitives for asynchronous events, and native support for serverless.
github.com/hydro-project/inf…
So excited to introduce Infinity, an agent runtime in Rust for extreme concurrency!
Infinity brings a new architecture that treats agent concurrency as a scheduling problem, tool call primitives for asynchronous events, and native support for serverless.
github.com/hydro-project/inf…
Infinity introduces fundamentally new forms of tool calls.
Tools can stream asynchronous events over time, which are revealed to the agent as “synthetic tool calls”, which inject the events in a way that preserves post-training capabilities.
This is an early research project, and we think there are lots of opportunities to combine our efforts with research around RLMs, context optimization, and more.
Learn more in our blog post! aws.amazon.com/blogs/opensou…
So excited to kick off tonight’s SF Systems with @mmalisper showing us how to rewrite Postgres in Rust! Thanks @recallai for hosting.
Shadaj Laddad retweeted
What do agents actually do most of the time?
They wait: for inference, tool calls, input.
What if instead they did suspend & resume to eliminate idle time
@ShadajL announced Infinity today at AGNTCon, an agent runtime where scale to zero is the default
infinity.hydro.run/
Shadaj Laddad retweeted
Amazing presentation by @ShadajL at AGNTCon!
Shadaj Laddad from @awscloud is tackling exactly that at AGNTCon + MCPCon NA with “Predictable Correctness for Unpredictable Code.”
The goal? Use Hydro and Rust to catch distributed systems errors at compile time, even when agents write the code.
Catch the keynote live: bit.ly/4dVkjdO
Shadaj Laddad retweeted
Interesting talk from @ShadajL on how his AWS team enforces universal properties on their distributed systems and agents with two Rust frameworks: Hydro.run and infinity.hydro.run
His tip: discover your own universal properties and write tools to enforce them.
It was awesome to share our work at @awscloud with the Agentic AI community!
Shadaj Laddad from @awscloud is tackling exactly that at AGNTCon + MCPCon NA with “Predictable Correctness for Unpredictable Code.”
The goal? Use Hydro and Rust to catch distributed systems errors at compile time, even when agents write the code.
Catch the keynote live: bit.ly/4dVkjdO
The SF Systems Club is back with an evening of talks on the future of data, featuring talks from @rescrv on agents-only version control and @mmalisper on rewriting Postgres in Rust. Thanks @recallai for hosting!
luma.com/osts92ax
Shadaj Laddad retweeted
New blog post!
Last week, I set out to build the best 2B-sized decision model (what the TypeSafe AI folks call "system one" models, like Jev). I succeeded, for about 24 hours before being overtaken in the benchmarks.
And now @sh_reya is presenting the world premiere of Quail, a domain-specific LLM inference engine for AI-SQL!
SF Systems Meetup off to an incredible talk with @parker_ziegler’s talk on how we can bring the power of programming languages to mediums beyond text!
This is such a cool project, and makes me really curious what interface an inference engine should expose beyond naive input / output. Precise control over schedules, caching, etc will be key.
Also, excited to learn more in Shreya’s talk tonight at SF Systems!
Shadaj Laddad retweeted
excited to see everyone tomorrow night! sorry this one “sold out”
if you are interested in hosting in the future (and have a large space) - let us know
SF Systems is back with a night of talks from researchers bringing their work to the real world!
@sh_reya will share her latest work on domain-specific inference engines and @parker_ziegler will show us what programming can look like beyond text!
luma.com/mzkxb97z
It’s time to get formal
github.com/hydro-project/hyd…
Based on the awesome research from @conor_power23’s PhD dissertation, which showed that we can harness formal verification for distributed systems without trusting LLMs to write correct specs!
