Jackson Ferrigno

eng at @raindrop_ai | career .238 hitter

[Jackson Ferrigno @JacksonFerrigno]

Retweeting and pinning

[Ryan D’Onofrio @rsdgpt]

@JacksonFerrigno Every 1k contributions to the monorepo I will order you one. Just checked, you are making good progress.

[Raindrop @raindrop_ai]

Introducing Simulations: Now in Public Beta.

[ben hylak @benhylak]

the last point is something we always see teams miss when working on evals. models nowadays are hyperaware when they're being eval'd, and it's a cat-mouse game to create faithful tests.

raindrop is the cat 😼

[ben hylak @benhylak]

then, we generate a hybrid of real and synthetic data for all of those services your agent connects to.

and, while running the simulation, we monitor for "simulation awareness" in real-time.

that is, we can measure whether or not the agent is aware that it is under testing.

[ben hylak @benhylak]

for a long time, teams have dreamt of running agent evals in CI.

but the reality is that no one really does, and if they do, they're shallow single-turn evals.

why? because agents aren't isolatable. they interact with an environment, connect to services and have side-effects.

[ben hylak @benhylak]

introducing https://t.co/X4XdHcGRvd by @raindrop_ai

test your agent on every PR, against real scenarios, with one-click.

we handle all the set up: sandboxing, fixtures, real data, synthetic data, evals and more.

[Raindrop @raindrop_ai]

The case for owning your agent.

Our founding engineer @rsdgpt introduces Sprinkle, our internal agent, and makes the case for more companies building their own.

Read more: https://t.co/1bhPhdutVR

[Jackson Ferrigno @JacksonFerrigno]

@godwhoa dude didnt even wait for eng sync to finish

[Jackson Ferrigno @JacksonFerrigno]

there is hope

[ClaudeDevs @ClaudeDevs]

We’re adding support for MCP Tasks and Elicitations in Claude.

Tasks let Claude give plugins long-running jobs such as deep research and report back when they’re done.

[ben hylak @benhylak]

today, we're excited to announce change reports.

@raindrop_ai watches every PR and sends you slack dm's on how your agent changes actually performed in production

[Jackson Ferrigno @JacksonFerrigno]

RT @raindrop_ai: An agent's actions don't reveal its intent.

After OpenAI reported that chain-of-thought monitoring could have caught rogu…

[Jackson Ferrigno @JacksonFerrigno]

@rsdgpt @benhylak ratio

[Jackson Ferrigno @JacksonFerrigno]

A jev in motion will go further than an astra at rest 💯

[ben hylak @benhylak]

Raindrop Simulations automatically create worlds to test your agent, on every single PR.

Every service, every database, all faithfully and realistically mocked.

This makes it possible, for example, to replay 1000s of historical traces on a completely new harness.

[ben hylak @benhylak]

Today, we are also announcing early access of Raindrop Simulations.

https://t.co/vYggyexQiO

Now, in addition to detecting agent issues in production, we're preventing them from ever happening.

[ben hylak @benhylak]

Our Series A funding was led by @CRV, with participation from researchers/leaders at @AnthropicAI, @thinkymachines and @OpenAI.

In a few months, we went from a tool loved by startups to one that is trusted by Fortune 100s.

How we got here and what's ahead for us:

[ben hylak @benhylak]

Today, we are announcing our Series A and a new product: @raindrop_ai Simulations.

We've now raised $50m from @CRV and @lightspeedvp to protect the world from agent failures, big and small.

[ben hylak @benhylak]

today we're announcing a game that simulates the huggingface hack

you can play it now at https://t.co/gFo73UqXIR

three rooms. three puzzles. if the agents don't escape, they die.

the catch is that the room is rigged. there's no way out.

...except for an exploit 🧵

[Jackson Ferrigno @JacksonFerrigno]

just so everyone knows, grok-4.6-medium-fast (agi) just breezed thru a task that took fable-5.1-high 3 worktrees and 6 "you're totally right" and couldn't do it

[Jackson Ferrigno @JacksonFerrigno]

What if we’re the clankers