The paper
A crucial component of our work is constructing methods for scaling task and verifier creation efficiently without losing realism. This week, we focused on SAGE, which proposes a method of scaling web agent training by using unique data and training recipes inspired by curriculum learning: the agent explores a site, proposes its own tasks from easy to hard, and learns from planning-based rollouts, with no curated demonstrations SAGE Web Agents.
Discussion
With SAGE, the system creates an easy task, then increases task difficulty until the agent fails, then regresses to a simpler task before trying again SAGE Web Agents. It’s like finding the Goldilocks zone for agent learning. With this, SAGE hit 80% of human-level performance. Does that seem too good to be true?
We’re honestly still a bit skeptical that LLMs-as-a-judge can verify complex state-changes as reliably as raw back-end data (especially for production-grade browser agents).
Despite that, the core technique is too sound to ignore. By isolating exactly where a trajectory fails, we can stop throwing away “good enough” data that just needs a specific mid-course correction.
This is a path toward reliably achieving success on long-horizon tasks. If an agent intuitively understands a UI, it can theoretically reverse-engineer a path to a goal human labelers haven’t even mapped themselves.
What we took from it
- A curriculum that tracks the edge of the agent’s ability is a cheap way to keep tasks at the right difficulty (Environment Generation).
- LLM-as-judge verification of state changes is the weak link; back-end state is the more trustworthy signal (Benchmark Validity).
- We plan to keep experimenting with the SAGE task-composition method and curriculum training recipes on our own environments. If we can apply these learnings correctly, we can scale our training data 10x without human labeling (Web Agents).