Teahose.
SIGN IN
NEW HERE — WHAT TEAHOSE DOES
We read the entire AI & tech firehose — so you don't have to.
PODPodcastsAll-In, No Priors, Acquired…
NEWNewslettersStratechery, Newcomer…
PAPPapersPhysical AI research
PHProduct Huntdaily launches
VCInvestor ScoutSequoia, a16z, Benchmark…
CLAUDE DISTILLS →
7 reads, 30 sec each — free, 6 AM ET.
+ a live graph of the companies, people & themes underneath.
HOME/PEOPLE/FEDERICO CASSANO
// PERSON

Federico Cassano

ROLE RESEARCH LEAD, COMPOSER 2MENTIONS 4LAST SEEN MAY 26, 2026
// BIO

Federico Cassano is a researcher and Research Lead at Cursor (Anysphere), where he leads development of Composer 2, the company's proprietary agentic coding model. He is best known for architecting a globally distributed reinforcement learning training system built in collaboration with Fireworks AI, which disaggregates training and inference across multiple clusters and syncs weight deltas rather than full model checkpoints. Prior to Cursor, he conducted academic research at Northeastern University's PRL lab under Arjun Guha and held positions at Scale AI, Roblox, and Trail of Bits.

// RECENT MENTIONS
// SIGNALS
4 SIGNALS
01
mention·Training Data·MAY 26, 2026

Every serious AI application company will eventually need to train its own models to compete. Prompt engineering and off-the-shelf models hit a ceiling — the real leverage is baking your application's specific behavior, tools, and environment directly into model weights.

Source
02
product·Training Data·MAY 26, 2026

Cursor and Fireworks built a globally distributed system where training and inference are disaggregated across multiple clusters — including repurposing production inference GPUs during off-peak hours — and syncing weight deltas (not full models) across the globe.

Source
03
product·Training Data·MAY 26, 2026

We started from a very strong base, which is Kimi 2.5. That's like a one trillion parameter MOE that's 30B active. So very, very sparse, actually.

Source
04
mention·Training Data·MAY 26, 2026

There is actually this kind of myth that during RL, you spend more, way more inference flops than training flops. This is sort of like just because the open source inference engines are very unoptimized instead of actually being a property of RL. Roughly the same ratio... In theory, if you push the GPUs to the maximum, you should have one third of your training GPUs allocated to inference.

Source

AI-extracted from podcast / newsletter / paper summaries. May contain errors.

Federico Cassano · Research Lead, Composer 2 — 4 mentions on Teahose