Eragon (@EragonAI) / X

Eragon

33 posts

Eragon

@EragonAI

The AI Operating System For Work

San Francisco

eragon.ai

Joined September 2025

Following

663

Followers

Eragon reposted
AgentMail (YC S25)
@agentmail
Jul 1
@EragonAI agents just got their own inbox, powered by AgentMail! They read email, pull what matters, and run the workflow. When something sensitive comes up, they email you to approve. Link in comments to learn more
768
Eragon
@EragonAI
Jul 1
Eragon agents now have their own email inbox through @agentmail Email lands → agent reads it → kicks off the workflow → emails you for approval before anything sensitive happens.
1.5K
Eragon reposted
Rishabh Tiwari
@rish2k1
Jun 20
Article
Why On-Policy Distillation Works and Naive Self-Distillation Doesn't
A dense training signal that points in the wrong direction. Authors: @rish2k1 @KushaSareen @Devvrit_Khatri @LakshyAAAgrawal @KurtKeutzer @matei_zaharia @inderjit_ml @agarwl_ TL;DR Reinforcement...
140K
Eragon
@EragonAI
Jun 16
Most enterprise AI fails not because models are bad — but because deployment is. Our 3 bets on what actually drives ROI: → Memory > raw intelligence → Right model for the task > one frontier model for everything → Forward-deployed engineers > plug-and-play platforms Read our
173
Eragon
@EragonAI
Jun 16
3 Thesis on AI Agents
From eragon.ai
151
Eragon reposted
Aaron Wang
@authenticdasein
Jun 9
Kinda crazy that @EragonAI just navigated through the entire DoorDash page via browser use and ordered us our favorite Oren's Hummus 😎🤑
00:00
1.5K
Eragon
@EragonAI
May 31
2/ A core issue with parameter-only RL is that it forces task-specific learning into the model weights. Traditional RL can improve model performance on the current task, but it also tends to shift behavior away from the base model, increase forgetting and reduce plasticity. On
246
Eragon
@EragonAI
May 31
4/ This reframes post-training. The default view treats adaptation as one channel — push every improvement into the weights — and pays for it with forgetting, eroded generality, and lost plasticity. FST splits that into two channels that co-evolve: task-specific nuance lives in
Eragon | Careers
From eragon.ai
196
Eragon
@EragonAI
May 30
3/ FST beats RL-only across four axes: - Data efficiency: FST reaches RL's running peak in substantially fewer optimizer steps — 3.0× fewer on CodeIO, 1.4× on Math (Polaris), and 3.0× on HoVer-hard — and continuing past the crossover, FST's running peak also exceeds RL's on all
154
Eragon reposted
Josh Sirota
@joshua_sirota
May 29
Replying to @joshua_sirota
How FST Works: To leverage the strong in-context learning of current LLMs, we treat the context as "fast weights" and model parameters as "slow weights", drawing from a rich literature in classic ML
827
Eragon reposted
Josh Sirota
@joshua_sirota
May 29
Announcing Fast-Slow Training (FST) pairing "slow" weights with "fast" context. We try to answer the question, can LLMs adapt continually without losing base skills? FST vs RL: - 3x more sample-efficient -Higher performance ceiling - Less KL drift - Continual learning: succeeds
1.2K
Eragon reposted
Dave Anderson
@MrDaveAllen
May 28
Excited to share our first research paper Learning, Fast and Slow: Towards LLMs That Adapt Continually. Fast-Slow Training (FST) combines optimized context with model weight updates. Read more here: arxiv.org/pdf/2605.12484
Eragon
@EragonAI
May 28
1/ At Eragon, we’re building an AI operating system that connects a company’s entire tech stack into a single interface for work, powered by a model post-trained on the customer’s own data so it understands the company’s unique context. We believe that AI system post-training
00:00
755
Eragon
@EragonAI
May 28
1/ At Eragon, we’re building an AI operating system that connects a company’s entire tech stack into a single interface for work, powered by a model post-trained on the customer’s own data so it understands the company’s unique context. We believe that AI system post-training
00:00
5.6K
Eragon
@EragonAI
May 28
Shoutout to researchers from @UCBerkeley @Mila_Quebec @UTAustin @periodiclabs and Mirendil on this collaboration!
503
Eragon
@EragonAI
May 28
Replying to @EragonAI
3/ FST beats RL-only across four axes: - Data efficiency: FST reaches RL's running peak in substantially fewer optimizer steps — 3.0× fewer on CodeIO, 1.4× on Math (Polaris), and 3.0× on HoVer-hard — and continuing past the crossover, FST's running peak also exceeds RL's on all
538
Eragon
@EragonAI
May 28
4/ This reframes post-training. The default view treats adaptation as one channel — push every improvement into the weights — and pays for it with forgetting, eroded generality, and lost plasticity. FST splits that into two channels that co-evolve: task-specific nuance lives in
Eragon | Careers
From eragon.ai
426